From b7a6b3cc8616bca25081f23d2d67fced0f1b2c90 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Thu, 30 Jul 2026 16:37:02 -0700 Subject: [PATCH 01/14] [[TRTLLM-13948][feat] Clean up DeepSeek tests using CPP Transceiver v1 Signed-off-by: Asfiya Baig --- .../accuracy/test_disaggregated_serving.py | 174 ++---------------- .../test_lists/qa/llm_function_core.txt | 6 - .../test_lists/qa/llm_function_rtx6k.txt | 4 - .../test_lists/test-db/l0_dgx_b200.yml | 2 - .../test_lists/test-db/l0_dgx_b300.yml | 1 - .../test_lists/test-db/l0_dgx_h100.yml | 2 - .../test_lists/test-db/l0_dgx_h200.yml | 4 - tests/integration/test_lists/waives.txt | 6 - 8 files changed, 17 insertions(+), 182 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 806d706cd5f1..a3e10df9f850 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1044,54 +1044,17 @@ class TestDeepSeekV3Lite(LlmapiAccuracyTestHarness): @pytest.mark.skip_less_device(2) @pytest.mark.skip_less_device_memory(60000) @skip_no_hopper - def test_nixl_backend(self): - ctx_server_config = { - "disable_overlap_scheduler": True, - "cache_transceiver_config": { - "backend": "NIXL", - "max_tokens_in_buffer": 4096 - } - } - gen_server_config = { - "disable_overlap_scheduler": True, - "cache_transceiver_config": { - "backend": "NIXL", - "max_tokens_in_buffer": 4096 - } - } - disaggregated_server_config = { - "hostname": "localhost", - "backend": "pytorch", - "context_servers": { - "num_instances": 1 - }, - "generation_servers": { - "num_instances": 1 - } - } - with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config, gen_server_config, - self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - - @pytest.mark.skip_less_device(2) - @pytest.mark.skip_less_device_memory(60000) - @skip_no_hopper - @pytest.mark.parametrize("transceiver_runtime", ["PYTHON", "CPP"], - ids=["python", "cpp"]) - def test_gen_only_sync(self, transceiver_runtime): - """Test gen-only synchronous KV transfer with each NIXL runtime. + def test_gen_only_sync(self): + """Test gen-only synchronous KV transfer with PYTHON NIXL runtime. Sets TRTLLM_DISABLE_KV_CACHE_TRANSFER_OVERLAP=1 so the gen worker calls - the blocking request_and_receive_sync path. The C++ variant uses a - bounded client timeout so a stuck transfer fails this test instead of - waiting for its outer one-hour timeout. + the blocking request_and_receive_sync path. """ ctx_server_config = { "disable_overlap_scheduler": True, "cache_transceiver_config": { "backend": "NIXL", - "transceiver_runtime": transceiver_runtime, + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 4096, }, } @@ -1099,7 +1062,7 @@ def test_gen_only_sync(self, transceiver_runtime): "disable_overlap_scheduler": True, "cache_transceiver_config": { "backend": "NIXL", - "transceiver_runtime": transceiver_runtime, + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 4096, }, } @@ -1120,10 +1083,8 @@ def test_gen_only_sync(self, transceiver_runtime): self.MODEL_PATH, # Apply to both servers: gen worker uses sync receive path. extra_env={"TRTLLM_DISABLE_KV_CACHE_TRANSFER_OVERLAP": "1"}, - request_timeout_s=(120 if transceiver_runtime == "CPP" else - DEFAULT_REQUEST_TIMEOUT_S), - request_max_retries=(0 - if transceiver_runtime == "CPP" else None), + request_timeout_s=DEFAULT_REQUEST_TIMEOUT_S, + request_max_retries=None, ) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) @@ -1160,47 +1121,6 @@ def test_gen_only_spec_dec(self): tensor_parallel_size=4) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - @pytest.mark.skip_less_device(8) - @parametrize_with_ids("overlap_scheduler", [True, False]) - @parametrize_with_ids("mtp_nextn", [0, 2]) - @pytest.mark.skip_less_device(8) - def test_auto_dtype(self, overlap_scheduler, mtp_nextn): - ctx_server_config = {"disable_overlap_scheduler": True} - gen_server_config = {"disable_overlap_scheduler": not overlap_scheduler} - ctx_server_config["cache_transceiver_config"] = { - "backend": "DEFAULT", - "max_tokens_in_buffer": 4096 - } - gen_server_config["cache_transceiver_config"] = { - "backend": "DEFAULT", - "max_tokens_in_buffer": 4096 - } - if mtp_nextn > 0: - ctx_server_config["speculative_config"] = { - "decoding_type": "MTP", - "max_draft_len": mtp_nextn - } - gen_server_config["speculative_config"] = { - "decoding_type": "MTP", - "max_draft_len": mtp_nextn - } - disaggregated_server_config = { - "hostname": "localhost", - "backend": "pytorch", - "context_servers": { - "num_instances": 1 - }, - "generation_servers": { - "num_instances": 1 - } - } - with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config, - gen_server_config, - self.MODEL_PATH, - tensor_parallel_size=4) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - @skip_pre_blackwell @pytest.mark.skip_less_device(8) @pytest.mark.parametrize("gen_pp,gen_tp,gen_cp,enable_attention_dp", [ @@ -1246,7 +1166,8 @@ def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config, "enable_chunked_prefill": False, "cuda_graph_config": None, "cache_transceiver_config": { - "backend": "UCX", + "backend": "NIXL", + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 8192, }, } @@ -1266,7 +1187,8 @@ def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config, "enable_chunked_prefill": False, "cuda_graph_config": cuda_graph_config, "cache_transceiver_config": { - "backend": "UCX", + "backend": "NIXL", + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 8192, }, "enable_attention_dp": enable_attention_dp, @@ -1299,7 +1221,8 @@ def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): }, "guided_decoding_backend": backend, "cache_transceiver_config": { - "backend": "DEFAULT", + "backend": "NIXL", + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 4096 } } @@ -1310,7 +1233,8 @@ def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): }, "guided_decoding_backend": backend, "cache_transceiver_config": { - "backend": "DEFAULT", + "backend": "NIXL", + "transceiver_runtime": "PYTHON", "max_tokens_in_buffer": 4096 } } @@ -1378,7 +1302,7 @@ def test_kv_cache_v2_nixl_python(self): with launch_disaggregated_llm(disaggregated_server_config, ctx_server_config, gen_server_config, self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) + run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) @pytest.mark.skip_less_device(4) @pytest.mark.skip_less_device_memory(60000) @@ -1676,70 +1600,6 @@ class TestDeepSeekV32Exp(LlmapiAccuracyTestHarness): MODEL_NAME = "deepseek-ai/DeepSeek-V3.2-Exp" MODEL_PATH = f"{llm_models_root()}/DeepSeek-V3.2-Exp-FP4-v2" - @pytest.mark.skip_less_device(8) - @pytest.mark.parametrize("overlap_scheduler", [False]) - def test_auto_dtype(self, overlap_scheduler): - cache_transceiver_config = { - "backend": "DEFAULT", - "max_tokens_in_buffer": 4096 - } - max_num_tokens = 8192 - ctx_kv_cache_config = { - "free_gpu_memory_fraction": 0.3, - "tokens_per_block": 64, - "dtype": "fp8", - } - moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} - ctx_server_config = { - "disable_overlap_scheduler": True, - "cuda_graph_config": None, - "cache_transceiver_config": cache_transceiver_config, - "kv_cache_config": ctx_kv_cache_config, - "tensor_parallel_size": 4, - "pipeline_parallel_size": 1, - "max_batch_size": 16, - "max_num_tokens": max_num_tokens, - "enable_autotuner": False, - } - gen_kv_cache_config = { - "free_gpu_memory_fraction": 0.5, - "tokens_per_block": 64, - "dtype": "fp8", - } - gen_server_config = { - "disable_overlap_scheduler": overlap_scheduler, - "cuda_graph_config": None, - "cache_transceiver_config": cache_transceiver_config, - "kv_cache_config": gen_kv_cache_config, - "moe_config": moe_config, - "max_batch_size": 128, - "max_num_tokens": 1024, - "cuda_graph_config": None, - "tensor_parallel_size": 4, - "pipeline_parallel_size": 1, - "moe_expert_parallel_size": 4, - "enable_attention_dp": True, - "enable_autotuner": False, - } - disaggregated_server_config = { - "hostname": "localhost", - "backend": "pytorch", - "context_servers": { - "num_instances": 1 - }, - "generation_servers": { - "num_instances": 1 - } - } - with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config=ctx_server_config, - gen_server_config=gen_server_config, - model_name=self.MODEL_PATH, - max_workers=128) as llm: - run_accuracy_test(llm, - model_name=self.MODEL_NAME, - test_sets=["MMLU", "GSM8K"]) - @pytest.mark.skip_less_device(4) @pytest.mark.skip_less_device_memory(200000) @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], @@ -1794,7 +1654,7 @@ def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): with launch_disaggregated_llm(disaggregated_server_config, ctx_server_config, gen_server_config, self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["GSM8K"]) + run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) @pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 9fde8bb9d1fa..41266d26832a 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,10 +1,5 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python] @@ -13,7 +8,6 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1] diff --git a/tests/integration/test_lists/qa/llm_function_rtx6k.txt b/tests/integration/test_lists/qa/llm_function_rtx6k.txt index 4260a5328d77..273f79ce5b11 100644 --- a/tests/integration/test_lists/qa/llm_function_rtx6k.txt +++ b/tests/integration/test_lists/qa/llm_function_rtx6k.txt @@ -1,7 +1,3 @@ -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml index cd8ab8f2ae74..a8a3beb86535 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml @@ -17,7 +17,6 @@ l0_dgx_b200: tests: - unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy - accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[cpp] # ------------- KV Cache V2 Scheduler IT (multi-GPU) --------------- - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens @@ -183,7 +182,6 @@ l0_dgx_b200: - accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4_mtp_index_share[tp_size=8-ep_size=8] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy TIMEOUT (240) - examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke TIMEOUT (120) - - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] TIMEOUT (60) - accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4 TIMEOUT (180) - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4ProDSpark::test_gsm8k_dep8_megamoe_deepgemm TIMEOUT (240) diff --git a/tests/integration/test_lists/test-db/l0_dgx_b300.yml b/tests/integration/test_lists/test-db/l0_dgx_b300.yml index 736319d8409f..0d2a254c67af 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b300.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b300.yml @@ -69,7 +69,6 @@ l0_dgx_b300: - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1] diff --git a/tests/integration/test_lists/test-db/l0_dgx_h100.yml b/tests/integration/test_lists/test-db/l0_dgx_h100.yml index ff1642f5e1e4..3feea471aaeb 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_h100.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_h100.yml @@ -34,8 +34,6 @@ l0_dgx_h100: - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python] - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ngram - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python diff --git a/tests/integration/test_lists/test-db/l0_dgx_h200.yml b/tests/integration/test_lists/test-db/l0_dgx_h200.yml index 3b40b905220b..2590c7f90b12 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_h200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_h200.yml @@ -27,10 +27,6 @@ l0_dgx_h200: - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8] - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3] - accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp2] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp2] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=1-ctx_pp=4] diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 473eb870caf4..c8883de9e597 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -248,8 +248,6 @@ full:GB300/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8_moe_d full:GB300/accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized SKIP (https://nvbugs/6479708) full:GB300/accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3_5_35B_A3B_VL::test_fp8_prequantized SKIP (https://nvbugs/6526429) full:GB300/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) -full:H100/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6313072) -full:H100/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6313072) full:H100/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=False] SKIP (https://nvbugs/6422343) full:H100/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=True] SKIP (https://nvbugs/6422343) full:H100/accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format SKIP (https://nvbugs/6528834) @@ -263,8 +261,6 @@ full:H100/disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_t full:H100/disaggregated/test_disaggregated.py::test_disaggregated_stress_test[input8k-output1k-conc512-qwen3_32b_fp8_stress] SKIP (https://nvbugs/6312828) full:H100/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) full:H100_PCIe/unittest/llmapi/test_llm_pytorch.py::test_llama_7b_multi_lora_evict_and_reload_lora_gpu_cache SKIP (https://nvbugs/5682551) -full:H20/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6345827) -full:H20/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6345827) full:H20/accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_fp8] SKIP (https://nvbugs/6327718) full:H20/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=False] SKIP (https://nvbugs/6422343) full:H20/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=True] SKIP (https://nvbugs/6422343) @@ -298,8 +294,6 @@ full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_ full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False] SKIP (https://nvbugs/6313076) full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash SKIP (https://nvbugs/6273850) full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8 SKIP (https://nvbugs/6273850) -full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6388153) -full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6400067) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] SKIP (https://nvbugs/6473374) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] SKIP (https://nvbugs/6388153) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True] SKIP (https://nvbugs/6313072) From 84eb8ab3aeafb1bf04725e719ae2b8f616872d33 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 14:52:17 -0700 Subject: [PATCH 02/14] remove test_kv_cache_v2_nixl_python and keep test_auto_dtype for more coverage of DSV3.2 Signed-off-by: Asfiya Baig --- .../accuracy/test_disaggregated_serving.py | 81 ++++++++++--------- .../test_lists/qa/llm_function_core.txt | 2 +- .../test_lists/test-db/l0_dgx_b200.yml | 1 + .../test_lists/test-db/l0_dgx_b300.yml | 1 - 4 files changed, 47 insertions(+), 38 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index a3e10df9f850..e79dfe4b7318 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1600,61 +1600,70 @@ class TestDeepSeekV32Exp(LlmapiAccuracyTestHarness): MODEL_NAME = "deepseek-ai/DeepSeek-V3.2-Exp" MODEL_PATH = f"{llm_models_root()}/DeepSeek-V3.2-Exp-FP4-v2" - @pytest.mark.skip_less_device(4) - @pytest.mark.skip_less_device_memory(200000) - @pytest.mark.parametrize("use_kv_cache_manager_v2", [False], - ids=["cache_mgr_v1"]) - def test_kv_cache_v2_nixl_python(self, use_kv_cache_manager_v2): - """Test with KV cache manager v1, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + @pytest.mark.skip_less_device(8) + @pytest.mark.parametrize("overlap_scheduler", [False]) + def test_auto_dtype(self, overlap_scheduler): + cache_transceiver_config = { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 + } max_num_tokens = 8192 + ctx_kv_cache_config = { + "free_gpu_memory_fraction": 0.3, + "tokens_per_block": 64, + "dtype": "fp8", + } moe_config = {"backend": "TRTLLM", "max_num_tokens": max_num_tokens} ctx_server_config = { "disable_overlap_scheduler": True, - "kv_cache_config": { - "free_gpu_memory_fraction": 0.5, - "enable_block_reuse": False, - "use_kv_cache_manager_v2": use_kv_cache_manager_v2 - }, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 4096 - }, - "tensor_parallel_size": 2, - "moe_expert_parallel_size": 2, + "cuda_graph_config": None, + "cache_transceiver_config": cache_transceiver_config, + "kv_cache_config": ctx_kv_cache_config, + "tensor_parallel_size": 4, + "pipeline_parallel_size": 1, + "max_batch_size": 16, + "max_num_tokens": max_num_tokens, "enable_autotuner": False, } + gen_kv_cache_config = { + "free_gpu_memory_fraction": 0.5, + "tokens_per_block": 64, + "dtype": "fp8", + } gen_server_config = { - "disable_overlap_scheduler": False, + "disable_overlap_scheduler": overlap_scheduler, + "cuda_graph_config": None, + "cache_transceiver_config": cache_transceiver_config, + "kv_cache_config": gen_kv_cache_config, "moe_config": moe_config, - "kv_cache_config": { - "free_gpu_memory_fraction": 0.5, - "enable_block_reuse": False, - "use_kv_cache_manager_v2": use_kv_cache_manager_v2 - }, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 4096 - }, - "tensor_parallel_size": 2, - "moe_expert_parallel_size": 2, + "max_batch_size": 128, + "max_num_tokens": 1024, + "cuda_graph_config": None, + "tensor_parallel_size": 4, + "pipeline_parallel_size": 1, + "moe_expert_parallel_size": 4, + "enable_attention_dp": True, "enable_autotuner": False, } disaggregated_server_config = { "hostname": "localhost", "backend": "pytorch", "context_servers": { - "num_instances": 1, + "num_instances": 1 }, "generation_servers": { - "num_instances": 1, + "num_instances": 1 } } with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config, gen_server_config, - self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) + ctx_server_config=ctx_server_config, + gen_server_config=gen_server_config, + model_name=self.MODEL_PATH, + max_workers=128) as llm: + run_accuracy_test(llm, + model_name=self.MODEL_NAME, + test_sets=["MMLU", "GSM8K"]) @pytest.mark.timeout(DEFAULT_TEST_TIMEOUT) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 41266d26832a..aec114562a42 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,5 +1,5 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] -accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] +accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml index a8a3beb86535..762bda05cdf6 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml @@ -183,6 +183,7 @@ l0_dgx_b200: - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy TIMEOUT (240) - examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke TIMEOUT (120) - accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4 TIMEOUT (180) + - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4ProDSpark::test_gsm8k_dep8_megamoe_deepgemm TIMEOUT (240) - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp TIMEOUT (60) diff --git a/tests/integration/test_lists/test-db/l0_dgx_b300.yml b/tests/integration/test_lists/test-db/l0_dgx_b300.yml index 0d2a254c67af..ee04247921d4 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b300.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b300.yml @@ -68,7 +68,6 @@ l0_dgx_b300: - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1] - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1] From f8381b0839c8ad41551518dba3d4ddfbf8bcfbf0 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 15:01:55 -0700 Subject: [PATCH 03/14] remove TestDeepSeekV3Lite::test_auto_dtype_with_helix as helix can't run with Python transceiver Signed-off-by: Asfiya Baig --- tests/integration/defs/.test_durations | 3073 ++++++++--------- .../accuracy/test_disaggregated_serving.py | 87 - .../test_lists/qa/llm_function_core.txt | 2 - .../test_lists/test-db/l0_dgx_b200.yml | 3 +- tests/integration/test_lists/waives.txt | 3 + 5 files changed, 1504 insertions(+), 1664 deletions(-) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 5dbb75d922a7..92d9aca24bd1 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -1,1575 +1,1502 @@ { - "accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 563.5671428571428, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False]": 577.3825454545454, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 762.6861428571428, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False]": 248.946875, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True]": 155.819, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False]": 226.137625, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True]": 155.884, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2]": 274.7764117647059, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[adp-mtp2]": 309.67263636363634, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[noadp-mtp0]": 459.758, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_spec_dec": 174.414125, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python]": 261.8368118811881, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0]": 164.20842857142858, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2]": 373.866, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 141.08328571428572, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 176.04781981981984, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python": 283.61367924528304, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_nixl_backend": 256.39867676767676, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV4Flash::test_auto_dtype": 653.3680224719101, - "accuracy/test_disaggregated_serving.py::TestGLM52NVFP4::test_nvfp4_nixl_python[cache_mgr_v1]": 1014.7549897959184, - "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 337.72714285714284, - "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 431.009, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False]": 250.61634653465347, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True]": 243.31274747474748, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 247.33391089108912, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 423.1184242424243, - "accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4": 1168.2368000000001, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-False]": 205.6069387755102, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-True]": 205.10983333333334, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-False]": 213.14781632653063, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-True]": 213.95441836734696, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-False]": 164.90764646464646, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-True]": 162.11889795918367, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-False]": 176.48985294117648, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-True]": 175.97751515151515, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-False]": 204.73766336633665, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-True]": 202.2888383838384, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-False]": 216.0393039215686, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-True]": 214.77772549019608, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-False]": 163.57239603960394, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-True]": 161.92363725490196, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-False]": 176.5338673469388, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-True]": 173.95371287128714, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_beam_search": 819.4241428571429, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=1-ctx_pp=2]": 158.725, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=2-ctx_pp=2]": 133.752, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[MMLU-gen_tp=1-ctx_pp=2]": 85.508, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[MMLU-gen_tp=2-ctx_pp=2]": 85.066, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_eagle3[eagle3_one_model=True-overlap_scheduler=True]": 168.67477142857143, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_gen_only_spec_dec": 211.36951485148515, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 77.02336633663367, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 157.4601919191919, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python": 504.60869444444444, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[GSM8K]": 176.20839999999998, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[MMLU]": 108.71153, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp1pp2]": 198.364, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp1]": 138.52641836734696, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp1pp2]": 86.348, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp1]": 91.4269603960396, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=False]": 522.0529894736842, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=True]": 626.4120731707317, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=3-block_reuse=True-use_py_transceiver=False]": 476.09568085106383, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_ctx_dp2_gen_tp4": 589.0531538461539, - "accuracy/test_disaggregated_serving.py::TestQwen3NextInstruct::test_auto_dtype[use_py_transceiver=False]": 438.7, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False]": 341.3867755102041, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-True]": 342.382137254902, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[True-True]": 320.5173163265306, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2]": 296.2844343434343, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill": 340.61881372549016, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first": 1049.90698, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first_kv_cache_v1": 129.3505, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend": 174.3157075471698, - "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_fp8]": 254.05641666666665, - "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_nvfp4]": 325.10448793565683, - "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[qwen3vl_2b_instruct]": 181.72481818181816, - "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[no_overlap]": 66.90779939209726, - "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[overlap]": 66.4708875, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-ep2]": 192.40157142857143, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-tp2]": 294.607, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b]": 153.4064285714286, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[20b]": 131.14983333333333, - "accuracy/test_llm_api_autodeploy.py::TestGemma4MoE::test_bf16": 1030.2359108910891, - "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma3n_e2b_it": 662.0174712990937, - "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma4_e2b_it": 347.1145662650602, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_attention_dp[4]": 260.8819504950495, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[torch-True-1]": 928.2698, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[triton-False-1]": 297.40318934911244, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-1]": 151.84377401129942, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-4]": 130.9596782178218, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-True-1]": 297.85671428571425, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[flashinfer]": 194.51192857142857, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[trtllm]": 159.54107079646016, - "accuracy/test_llm_api_autodeploy.py::TestMiniMaxM2::test_finegrained_fp8": 580.2765643564356, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[Qwen_QwQ-32B-False]": 269.0364285714286, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[deepseek-ai_DeepSeek-R1-0528-True]": 802.7421428571429, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[google_gemma-3-1b-it-False]": 58.604, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.1-8B-Instruct-False]": 47.470142857142854, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.3-70B-Instruct-False]": 140.37328571428571, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Codestral-22B-v0.1-False]": 87.76414285714286, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Ministral-8B-Instruct-2410-False]": 61.62542857142857, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_DeepSeek-R1-0528-NVFP4-v2-True]": 1478.76, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-FP8-True]": 107.06828571428571, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-NVFP4-True]": 125.27557425742575, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-flashinfer_ssm-False]": 438.54785714285714, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-False]": 356.0815714285714, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-True]": 476.00142857142856, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-1-attn_dp_off-trtllm]": 1290.0525714285714, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 347.11914285714283, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-1-attn_dp_off-trtllm]": 389.7956233951498, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 260.72430348258706, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 219.20380198019802, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 298.9254, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 638.307, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 476.6877425742574, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 389.805, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 742.2181908602151, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-4-attn_dp_on-trtllm]": 546.3941666666666, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[bf16]": 243.7910835579515, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[fp8]": 219.9212, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer]": 511.63059999999996, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm]": 392.782, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[fp8_ws4_80gb-trtllm]": 512.1759714285714, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[nvfp4_ws4_80gb-trtllm]": 527.9766666666666, - "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-4]": 1914.121405940594, - "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-8]": 2471.7431428571426, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[False]": 394.80692307692306, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[True]": 528.0706153846154, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_fp8[True]": 503.2507582260372, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_nvfp4[True]": 415.859, - "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_nvfp4[8]": 1174.774, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[latency]": 2025.141625, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput]": 703.5015340136055, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency]": 814.2682857142856, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp]": 405.60157142857145, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp_tp4]": 750.1278415841583, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen]": 803.0506666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen_adp_lmtp]": 594.0771666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput]": 484.942600660066, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_bs8_mtp]": 616.6849142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp]": 431.8518361774744, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp4]": 786.3406, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8]": 425.4071666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_dsa_host_cache_offload[host_cache_offload]": 675.9716666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline]": 708.164, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline_mtp1]": 873.1390857142858, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_attn_multi_gpus": 620.9094285714285, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline]": 385.06701923076923, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline_mtp1]": 403.80720952380955, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[baseline]": 350.41063636363634, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3::test_skip_softmax_attention_multi_gpus[target_sparsity_0.9]": 1907.8475841584159, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 110.22128571428571, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 95.23816666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 112.59433333333332, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 117.33757142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 108.42275, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 96.67157142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 99.13671428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 104.493, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 136.719875, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 135.1635, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 93.43057142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 96.45733333333332, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 115.44271428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 108.53066666666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 173.38052512562814, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 240.23762278481013, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 89.461, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 98.76371428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 109.929625, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 116.24385714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 87.24416666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 89.3735, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 171.456, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 178.44, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 89.22883333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 93.58716666666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 233.06353576437587, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=False]": 490.6032997416021, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=True]": 672.0208601583113, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp": 81.99830310880829, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 74.9602857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 157.13757142857142, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 50.04, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 53.55142857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.368857142857145, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 52.22757142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.32028571428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 43.58842857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 124.31714285714285, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 37.46957142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 286.8615555555556, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[pp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 361.54089473684206, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 51.500142857142855, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 47.665571428571425, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.233, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 49.38442857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 50.68985714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 241.726, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 49.91157142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 60.44242857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 72.28114285714285, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 185.46715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 54.29342857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 150.5922857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 44.819714285714284, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 53.9998, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 95.5758, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.01414285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 54.855714285714285, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 46.03628571428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 70.931, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 49.37942857142857, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 69.58571428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 177.16992307692306, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 39.854571428571425, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 64.18114285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=0]": 110.4788, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=2]": 65.1326, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=0]": 57.580400000000004, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=2]": 68.72460000000001, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=0]": 73.8476, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=2]": 74.2298, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=False]": 204.60704605263157, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=True]": 317.57645359477124, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 242.8417142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 228.30883333333335, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 1257.124142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 282.825, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 334.87314285714285, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 826.2921666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=fp8-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 450.24012227074235, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]": 251.18992151162792, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 233.96449426934097, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 228.73876377952757, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 293.3109759358289, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_dummy_load_format": 185.64599129172714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 125.61314285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 141.97885714285712, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 92.35671428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 100.544, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.229, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 114.00025, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.495125, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 202.64102366863904, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 100.6405, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.03385714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 95.19957142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.24828571428571, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.07571428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.91071428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 203.96484821428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 100.222, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 128.5577142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.65114285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 48.82, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 132.92722115384615, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 89.69357142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.26528571428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 63.23810576923076, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 61.52645192307692, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[pp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 58.106, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[pp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 60.694, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 94.36728571428571, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.680571428571426, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 43.60585714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 113.014, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 50.377, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 73.366, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 51.053, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 50.206857142857146, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 49.96328571428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 69.58657142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 65.82942857142856, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 76.64985714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 59.55085714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 59.55585714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.456857142857146, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-sampler_async_worker=False]": 81.1257142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.917857142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 68.30914285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 130.82932673267325, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 61.337153846153846, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 58.744692307692304, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 110.86007920792079, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=True]": 114.5420297029703, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 48.49171428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 58.338857142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 50.66985714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.70885714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 129.97968316831683, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=0]": 65.82992835820896, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=2]": 146.14830721003133, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=0]": 51.27741747572816, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=2]": 58.42619417475728, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 74.263375, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 229.74792266666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]": 106.62016666666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=2]": 64.91603883495145, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=fp8-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 142.49904672897196, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 257.45325974025974, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 197.461, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 238.9005, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 129.17557142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 144.1535714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 118.58166666666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 125.60383333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 121.43585714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 137.706, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 126.48357142857144, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 136.42479999999998, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 118.6963217158177, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 168.66416666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 176.74114285714288, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 315.0251083123425, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 123.97833333333332, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 94.24726816380449, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 198.32728571428572, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 206.059, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 188.77516666666665, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 357.6437142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 413.533, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 235.61871428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 142.03583333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 204.68985714285714, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-pp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 264.8346666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 229.01957142857142, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 369.8146712962963, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 159.7267619047619, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 147.81376923076922, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 84.19792307692308, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 100.28721428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 214.388625, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 125.55446153846154, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 59.623, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 174.7900666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 169.331, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 72.06583333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 134.51646666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 87.48892592592593, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 151.09442, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 159.31716666666665, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 95.5052197309417, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 113.31939285714286, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 155.69638461538463, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 145.39335, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 146.07366666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 130.40482805429863, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 85.24828571428571, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 179.83242857142858, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 183.90396666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 77.26146728971963, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 261.2648656716418, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 114.26407692307693, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=True-moe_backend=TRTLLM]": 188.3487525773196, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_batch_waiting[batch_wait_timeout_iters=10-batch_wait_max_tokens_ratio=1.0-mtp_nextn=0-fp8kv=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-v2_kv_cache=True]": 222.85041397849463, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_auto_dtype": 790.9375, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=MEGAMOE_DEEPGEMM]": 559.4773168316832, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=TRTLLM]": 327.3962673267327, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy": 1220.98871875, - "accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4_mtp_index_share[tp_size=8-ep_size=8]": 801.905, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dflash": 606.7177058823529, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format": 82.15868983957219, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[cutlass-one_model-overlap_scheduler]": 1810.228, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[triton-one_model-overlap_scheduler]": 1204.1007378640775, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-no_overlap_scheduler]": 597.1911333333333, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-overlap_scheduler]": 446.03928155339804, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-no_overlap_scheduler]": 229.9775, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-overlap_scheduler]": 411.70846521739134, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-no_overlap_scheduler]": 474.76293333333336, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-overlap_scheduler]": 644.7451866028708, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-no_overlap_scheduler]": 397.8118333333333, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-overlap_scheduler]": 505.294358490566, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-one_model]": 259.8957142857143, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-two_model]": 324.8745714285714, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v2_kv_cache-one_model]": 708.3871666666666, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-cutlass-auto]": 238.89962115127173, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-auto]": 283.65757142857143, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-fp8]": 199.45164247311828, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-cutlass-auto]": 173.85969541778977, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-auto]": 201.94083333333333, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-fp8]": 315.4943882978723, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-triton-auto]": 1349.77619, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-auto]": 536.0184, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-fp8]": 459.6693106796116, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-triton-auto]": 512.76651, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-trtllm-auto]": 446.99943925233646, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-triton-auto]": 456.39875, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-trtllm-auto]": 544.3978755980862, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-triton-auto]": 844.6485600000001, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-auto]": 642.8702, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-fp8]": 573.480285, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-triton-auto]": 632.22521, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-auto]": 478.0725, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-fp8]": 326.0081666666667, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-tp4-trtllm-auto]": 536.486418079096, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache_no_reuse-tp4-cutlass-auto]": 412.47744, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus_online_eplb[fp8]": 276.85751871657754, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]": 331.16407, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype": 194.39304497751124, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_reuse": 119.92895819935691, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_without_reuse": 130.76036677115988, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse": 47.05186774193549, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_disable_overlap_scheduler": 46.59689393939394, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_no_partial_reuse": 61.084156050955414, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_partial_reuse": 62.60442056074766, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_partial_reuse": 44.13613738019169, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse": 43.938835913312694, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_disable_overlap_scheduler": 48.01080909090909, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_low_memory_available": 56.98198746081505, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=False]": 180.3642857142857, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=True]": 207.9544285714286, - "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_auto_dtype": 895.99875, - "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_fp8_prequantized": 418.2255, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8]": 439.1708333333333, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8]": 679.1593333333334, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8]": 974.601006779661, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8_attn_dp]": 783.5843333333333, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_bf16": 336.4756683291771, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_fp8": 246.60832780082987, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_nvfp4": 206.84562613195345, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_bf16_dflash": 310.8462520547945, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_fp8_dflash": 304.4109135802469, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_nvfp4_dflash": 303.8202086720867, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4": 147.18037917222964, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_4]": 117.7441439790576, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_64]": 124.89551358695653, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=False]": 96.55514285714285, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=True]": 128.05185714285716, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=False]": 78.98383333333332, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=True]": 113.277, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=FLASHINFER-torch_compile=False]": 119.661, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=FLASHINFER-torch_compile=False]": 39.89528571428572, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=False]": 38.87285714285714, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=True]": 45.632714285714286, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=False]": 94.90157575757576, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=True]": 68.93085714285715, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=TRTLLM-torch_compile=True]": 106.232, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=FLASHINFER]": 153.13166666666666, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=TRTLLM]": 48.5009577039275, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=FLASHINFER]": 164.974125, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=TRTLLM]": 59.59303960396039, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dflash": 165.76832544378698, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dummy_load_format": 18.441307692307692, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=False-eagle3_one_model=True-overlap_scheduler=True]": 976.3793964497042, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=True-eagle3_one_model=True-overlap_scheduler=True]": 971.4979968652037, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 67.56983333333334, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 109.339625, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 53.60133641975309, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 81.8483850931677, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 79.50657142857143, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 97.091, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 46.40235435435436, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 70.09814423076924, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[pp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 89.04066666666667, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 40.44528571428572, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 130.926, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 38.101142857142854, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 33.152, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 113.30497727272729, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 72.48057142857144, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 126.10539603960396, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 87.13407272727272, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 39.70928571428572, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 130.6536153846154, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 63.65173205741627, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 81.74710891089109, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[llguidance]": 46.254285714285714, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 48.451914373088684, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[llguidance]": 47.952, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[xgrammar]": 49.702, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[llguidance-eagle3_one_model=True]": 57.91375, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 45.978490797546016, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[llguidance]": 44.04633333333334, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[xgrammar]": 43.43625, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=False]": 62.13226470588235, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=True]": 78.61358743169399, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=True-attn_backend=TRTLLM-torch_compile=True]": 233.3722371967655, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=False]": 780.0762724458203, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=True]": 631.2467230769231, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B_Instruct_RocketKV::test_auto_dtype": 930.7815714285715, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp4_tp2pp2[torch_compile=True-enable_gemm_allreduce_fusion=False]": 827.042, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 671.1513591160221, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=True]": 653.0381851851852, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 743.1877736842105, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=True]": 747.0298399999999, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=True]": 1015.7231386138614, - "accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm]": 529.4998, - "accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm_eagle]": 764.3608, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_fp8": 301.294, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_nvfp4_marlin_multi_gpus[tp_size=8]": 409.00075, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-cpp_mamba_cache]": 376.7685, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-python_mamba_cache]": 308.124, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-cpp_mamba_cache]": 214.224, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-python_mamba_cache]": 215.894, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar": 383.92656190476185, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar_custom_op": 251.13214285714287, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_OFF]": 805.9374, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_ON]": 499.62153535353536, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[TEP4]": 782.1175999999999, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_hopper_w4a16": 535.0051428571428, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_online_eplb[moe_backend=TRTLLM]": 485.27279999999996, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 394.49414285714283, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp": 715.7141650485437, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3]": 948.6714285714286, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8]": 448.4485, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_block_reuse[ADP4_MTP]": 1103.1094324324324, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 994.767, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_static_eplb[moe_backend=CUTLASS]": 511.1553333333333, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_off-trtllm]": 555.4016868686869, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_on-cutedsl]": 673.2446534653466, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus": 2393.7385, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_2gpu_mtp_ar": 296.3352, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[dep4]": 403.4035, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[tep4]": 594.184, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-cutlass]": 310.09275, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-trtllm]": 288.0665, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[tp1_block_reuse-cutlass]": 330.1165, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]": 221.4197142857143, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp]": 265.43, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_cutlass]": 1021.0351428571429, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_trtllm_eagle3]": 325.0231538461539, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_dummy_load_format": 41.01788461538462, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=False]": 156.23925, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=True]": 189.51606918238994, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=False]": 122.25, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=True]": 143.73514285714288, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]": 122.8515, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=True]": 177.88014285714286, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=True]": 124.60616666666667, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=False]": 212.62992307692306, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=True]": 218.3575, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=False]": 203.124, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=True]": 169.98084864864865, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=False]": 258.0261875, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=True]": 286.8506956521739, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=False]": 201.9002142857143, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]": 211.51953271028037, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a16_mxfp4[latency-TRTLLM]": 77.09778494623656, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-CUTLASS]": 85.84242857142857, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-TRTLLM]": 90.60216981132075, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-CUTLASS]": 80.64685714285714, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-TRTLLM]": 76.41928421052631, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention[target_sparsity_0.9-fp8kv=True]": 813.8270840108402, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=False]": 709.772, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=True]": 554.5961538461539, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False]": 581.32, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=True]": 508.1515714285714, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-CUTLASS]": 275.259, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-TRTLLM]": 225.91433333333333, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-CUTLASS]": 235.185, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM]": 309.260198019802, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_mtp": 342.53534065934065, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False]": 438.6115, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=True]": 646.5637106918239, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_cutedsl]": 617.0652068965518, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_trtllm]": 461.029, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_block_reuse]": 632.7510101010101, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_trtllm]": 534.2323333333334, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 955.435089201878, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 662.74585, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_mtp3_gdn_replay_tep4": 526.8681475409836, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_bf16": 166.71264935064934, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash": 214.72679870129872, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8": 207.41428353658537, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8_piecewise_cuda_graph": 198.19608035714288, - "accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM]": 270.3366730245232, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dflash": 143.1653267973856, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dummy_load_format": 44.24248447204969, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=False]": 100.20931804281345, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=True]": 107.118, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=True-enable_draft_len_schedule=False]": 105.864125, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=True-enable_max_concurrency=False-enable_draft_len_schedule=False]": 194.1835, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales[latency]": 217.11881497480204, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales_early_first_token_response": 135.53371608832808, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[fp8-latency]": 38.061456464379944, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[mxfp8-latency]": 48.70666666666666, - "accuracy/test_llm_api_pytorch.py::TestSeedOss_36B::test_auto_dtype": 1182.1868333333332, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_cuda_graph_matches_eager_logits[tinyllama-1.1b]": 14.049864583333333, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[gemma-3-1b]": 18.696708333333333, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[phi-4-mini]": 17.96046875, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen2-7b]": 19.755377604166668, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen3-0.6b]": 8.566161458333333, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[starcoder2-3b]": 24.163671875, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[tinyllama-1.1b]": 6.4027109375, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_cuda_graph_matches_eager_logits[bert-yelp]": 6.059911458333333, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-cuda_graph]": 4.2082109375, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-eager]": 4.435528645833333, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_per_token_reward[qwen2.5-prm-7b]": 22.102708333333332, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-0.6b]": 11.882435443037975, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-8b]": 25.415469696969698, - "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized": 328.788006006006, - "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma4_26B_A4B::test_nvfp4": 345.3036666666667, - "accuracy/test_llm_api_pytorch_multimodal.py::TestMistralSmall24B::test_auto_dtype[forced_chunked_prefill]": 171.36042857142857, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[bf16]": 418.94666666666666, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8]": 651.2778228228228, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8_mmmu_encoder_cuda_graph]": 326.4243922651934, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]": 467.0641428571428, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype[forced_chunked_prefill]": 277.644, - "accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL::test_auto_dtype[forced_chunked_prefill]": 155.48857142857142, - "accuracy/test_llm_api_pytorch_ray.py::TestLlama3_1_8BInstruct::test_pp2_ray": 72.727, - "cpp/test_multi_gpu.py::test_cache_transceiver[2proc-ucx_kvcache-90]": 185.5999120879121, - "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-nixl_kvcache-90]": 248.42608791208792, - "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-ucx_kvcache-90]": 317.74205555555557, - "cpp/test_multi_gpu.py::test_fused_gemm_allreduce[4proc-90]": 21.271241758241757, - "cpp/test_multi_gpu.py::test_mpi_utils[90]": 231.0067802197802, - "cpp/test_multi_gpu.py::test_user_buffer[2proc-90]": 4.18345054945055, - "cpp/test_unit_tests.py::test_unit_tests[batch_manager-80]": 561.8357163323782, - "cpp/test_unit_tests.py::test_unit_tests[common-80]": 225.72357142857143, - "cpp/test_unit_tests.py::test_unit_tests[common-90]": 224.40657185628743, - "cpp/test_unit_tests.py::test_unit_tests[executor-80]": 130.4267142857143, - "cpp/test_unit_tests.py::test_unit_tests[kernels-80]": 943.2667142857143, - "cpp/test_unit_tests.py::test_unit_tests[kernels-90]": 731.6601886227545, - "cpp/test_unit_tests.py::test_unit_tests[layers-80]": 918.556, - "cpp/test_unit_tests.py::test_unit_tests[layers-90]": 794.5549492537313, - "cpp/test_unit_tests.py::test_unit_tests[runtime-80]": 678.604, - "cpp/test_unit_tests.py::test_unit_tests[thop-80]": 3.1942857142857144, - "cpp/test_unit_tests.py::test_unit_tests[thop-90]": 3.119434131736527, - "disaggregated/test_ad_disagg.py::test_async_eagle3_full_model_handoff": 97.08721782178218, - "disaggregated/test_ad_disagg.py::test_async_generation_matches_aggregate": 86.58474257425743, - "disaggregated/test_ad_disagg.py::test_async_generation_no_overlap_matches_aggregate": 84.61831683168317, - "disaggregated/test_ad_disagg.py::test_async_sharded_generation_handoff": 97.06769, - "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[deepseek_v3_mla]": 157.40455151515152, - "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[tinyllama]": 96.72484194528876, - "disaggregated/test_ad_disagg.py::test_disaggregated_logits[deepseek_v3_mla]": 56.42044545454546, - "disaggregated/test_ad_disagg.py::test_disaggregated_logits[tinyllama]": 34.52554819277108, - "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[deepseek_v3_mla]": 141.5973172205438, - "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[tinyllama]": 96.18908459214501, - "disaggregated/test_ad_disagg.py::test_tinyllama_batch_handoff_semantic_slots": 76.65452567975831, - "disaggregated/test_ad_disagg_trtllm_serve.py::test_openai_completion": 61.01671428571428, - "disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin]": 104.70482000000001, - "disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin]": 133.51976530612245, - "disaggregated/test_auto_scaling.py::test_minimal_instances[etcd-round_robin]": 103.74983673469387, - "disaggregated/test_auto_scaling.py::test_minimal_instances[http-round_robin]": 101.29590816326531, - "disaggregated/test_auto_scaling.py::test_service_discovery[etcd-round_robin]": 54.902530612244895, - "disaggregated/test_auto_scaling.py::test_service_discovery[http-kv_cache_aware]": 51.200849999999996, - "disaggregated/test_auto_scaling.py::test_service_discovery[http-round_robin]": 50.95837, - "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-load_balancing]": 205.29111224489796, - "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-round_robin]": 202.5159595959596, - "disaggregated/test_auto_scaling.py::test_worker_restart[http-load_balancing]": 201.06501999999998, - "disaggregated/test_disaggregated.py::test_disaggregated_benchmark_gen_only_insufficient_kv[TinyLlama-1.1B-Chat-v1.0]": 75.97949742268042, - "disaggregated/test_disaggregated.py::test_disaggregated_cache_aware_balance[TinyLlama-1.1B-Chat-v1.0]": 132.24700515463917, - "disaggregated/test_disaggregated.py::test_disaggregated_chat_completion_tool_calls[TinyLlama-1.1B-Chat-v1.0]": 71.68480506329114, - "disaggregated/test_disaggregated.py::test_disaggregated_conditional[TinyLlama-1.1B-Chat-v1.0]": 92.38360050251256, - "disaggregated/test_disaggregated.py::test_disaggregated_ctxpp2_genpp2[TinyLlama-1.1B-Chat-v1.0]": 71.839, - "disaggregated/test_disaggregated.py::test_disaggregated_ctxpp2_gentp2[TinyLlama-1.1B-Chat-v1.0]": 71.719, - "disaggregated/test_disaggregated.py::test_disaggregated_ctxtp2_genpp2[TinyLlama-1.1B-Chat-v1.0]": 69.536, - "disaggregated/test_disaggregated.py::test_disaggregated_cuda_graph[TinyLlama-1.1B-Chat-v1.0]": 80.77402295918367, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_conditional_v2[DeepSeek-V3-Lite-bf16]": 139.9444081632653, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_empty_batch[DeepSeek-V3-Lite-bf16]": 190.4600707070707, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 128.8671616161616, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 121.09470707070707, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8]": 118.99839393939394, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 99.04030303030302, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 136.5389226006192, - "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 72.53091237113402, - "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 180.8998811881188, - "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 92.5423762886598, - "disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0]": 83.46904984423675, - "disaggregated/test_disaggregated.py::test_disaggregated_mixed[TinyLlama-1.1B-Chat-v1.0]": 137.4564641025641, - "disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_test[req120-conc64-qwen3_32b_fp8_mixed_stress]": 199.7655, - "disaggregated/test_disaggregated.py::test_disaggregated_multi_gpu[TinyLlama-1.1B-Chat-v1.0]": 72.94412244897958, - "disaggregated/test_disaggregated.py::test_disaggregated_ngram[TinyLlama-1.1B-Chat-v1.0]": 87.9892202072539, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap[TinyLlama-1.1B-Chat-v1.0]": 100.85853292181069, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python[TinyLlama-1.1B-Chat-v1.0]": 80.67388775510204, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_bounce[TinyLlama-1.1B-Chat-v1.0]": 107.13556521739132, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_fabric_memory[TinyLlama-1.1B-Chat-v1.0]": 104.9951875, - "disaggregated/test_disaggregated.py::test_disaggregated_perf_metrics[TinyLlama-1.1B-Chat-v1.0]": 90.53827720207254, - "disaggregated/test_disaggregated.py::test_disaggregated_python_transceiver_host_offload[TinyLlama-1.1B-Chat-v1.0]": 72.30680000000001, - "disaggregated/test_disaggregated.py::test_disaggregated_qwen3_32b_fp8[Qwen3/Qwen3-32B-FP8]": 146.544875, - "disaggregated/test_disaggregated.py::test_disaggregated_single_gpu[TinyLlama-1.1B-Chat-v1.0]": 86.30057989690722, - "disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0]": 68.38216216216216, - "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer[False-TinyLlama-1.1B-Chat-v1.0]": 57.86059788359788, - "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer_missing_blocks[False-TinyLlama-1.1B-Chat-v1.0]": 60.099248, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_cancel_gen_requests[TinyLlama-1.1B-Chat-v1.0]": 34.60525471698113, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_llama_context_capacity[False-False-DeepSeek-V3-Lite-fp8/fp8]": 160.16489374999998, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[False-TinyLlama-1.1B-Chat-v1.0]": 49.2354213740458, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[True-TinyLlama-1.1B-Chat-v1.0]": 40.8620374251497, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[False-TinyLlama-1.1B-Chat-v1.0]": 47.87118947368421, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[True-TinyLlama-1.1B-Chat-v1.0]": 45.21462158054712, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8]": 99.36854098360655, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-True-DeepSeek-V3-Lite-fp8/fp8]": 78.65977846153847, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-False-DeepSeek-V3-Lite-fp8/fp8]": 127.02941666666668, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-True-DeepSeek-V3-Lite-fp8/fp8]": 114.31295031055902, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-False-TinyLlama-1.1B-Chat-v1.0]": 44.86971506105834, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-True-TinyLlama-1.1B-Chat-v1.0]": 48.939882653061225, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-False-TinyLlama-1.1B-Chat-v1.0]": 56.65890439276486, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-True-TinyLlama-1.1B-Chat-v1.0]": 58.08935368956743, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-False-Qwen3-8B-FP8]": 77.68070953101362, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-True-Qwen3-8B-FP8]": 70.43922955974844, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-False-Qwen3-8B-FP8]": 65.3021451104101, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-True-Qwen3-8B-FP8]": 68.98847094801222, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[False-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 56.87129813664596, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[True-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 69.73002752293579, - "disaggregated/test_workers.py::test_workers_conditional_disaggregation_deepseek_v3_lite_bf16[DeepSeek-V3-Lite-bf16]": 145.47861616161617, - "disaggregated/test_workers.py::test_workers_kv_cache_aware_router[TinyLlama-1.1B-Chat-v1.0]": 137.83586597938142, - "disaggregated/test_workers.py::test_workers_kv_cache_events[TinyLlama-1.1B-Chat-v1.0]": 78.49960406091371, - "examples/test_ad_guided_decoding.py::test_autodeploy_guided_decoding_main_json": 108.45550755287009, - "examples/test_ad_speculative_decoding.py::test_autodeploy_eagle3_one_model_acceptance_rate[flashinfer-torch-simple]": 102.75090207715134, - "examples/test_ad_speculative_decoding.py::test_eagle_model_with_weights": 2.719433234421365, - "examples/test_ad_speculative_decoding.py::test_nemotron_mtp_model_with_weights": 799.8663165680474, - "examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke": 954.67404, - "examples/test_llm_api_with_mpi.py::test_llm_api_single_gpu_with_mpirun[TinyLlama-1.1B-Chat-v1.0]": 47.71125796812749, - "examples/test_ray.py::test_llm_inference_async_ray": 126.3797485380117, - "examples/test_ray.py::test_llm_inference_distributed_ray[pp2]": 112.781, - "examples/test_ray.py::test_llm_inference_distributed_ray[tep2]": 158.26951485148516, - "examples/test_ray.py::test_llm_inference_distributed_ray[tp2]": 65.46412871287129, - "examples/test_ray.py::test_llm_inference_distributed_ray[tp2pp2]": 196.643, - "examples/visual_gen/test_visual_gen.py::test_cosmos3_example": 286.4563489583333, - "examples/visual_gen/test_visual_gen.py::test_flux1_example": 238.85635549872123, - "examples/visual_gen/test_visual_gen.py::test_flux1_lpips_against_golden": 61.382666666666665, - "examples/visual_gen/test_visual_gen.py::test_flux2_example": 370.0974792746114, - "examples/visual_gen/test_visual_gen.py::test_flux2_lpips_against_golden": 85.6677142857143, - "examples/visual_gen/test_visual_gen.py::test_flux2_reference_image_example": 289.24313513513516, - "examples/visual_gen/test_visual_gen.py::test_ltx2_cuda_graph_lpips_matches_eager": 293.57128571428575, - "examples/visual_gen/test_visual_gen.py::test_ltx2_example": 382.7444775725594, - "examples/visual_gen/test_visual_gen.py::test_ltx2_lpips_against_golden": 189.37857142857143, - "examples/visual_gen/test_visual_gen.py::test_qwen_image_example": 224.7498372093023, - "examples/visual_gen/test_visual_gen.py::test_qwenimage_cuda_graph_lpips_against_golden": 79.9952, - "examples/visual_gen/test_visual_gen.py::test_qwenimage_lpips_against_golden": 86.78375, - "examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough": 239.43309943181816, - "examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart": 198.98814145658264, - "examples/visual_gen/test_visual_gen.py::test_wan21_t2v_lpips_against_golden": 36.528571428571425, - "examples/visual_gen/test_visual_gen.py::test_wan22_t2v_lpips_against_golden": 63.1655, - "examples/visual_gen/test_visual_gen.py::test_wan_t2v_example": 571.5687341772151, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[attn2d_2x2]": 953.7801428571429, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[cfg2_ulysses2]": 270.7155, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[ulysses4]": 1041.747, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[cfg2_tp2]": 151.057, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2]": 73.04014285714285, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_attn2d_2x1]": 275.8075, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_ulysses2]": 181.01357142857142, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_changed_final_token_reuse_cuda_graph[v1]": 20.15875, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_changed_final_token_reuse_cuda_graph[v2]": 29.40618181818182, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_context_logits_after_final_token_reuse[v1]": 11.5116, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_context_logits_after_final_token_reuse[v2]": 16.64923076923077, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph[v1]": 9.9382, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph[v2]": 15.367846153846154, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph_tp2[v1]": 32.907, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph_tp2[v2]": 40.815, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_guided_decoding_after_final_token_reuse[v1]": 10.742833333333333, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_guided_decoding_after_final_token_reuse[v2]": 15.2503125, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_zero_runtime_draft_speculation_after_final_token_reuse[v1]": 21.316636363636363, - "kv_cache/test_final_single_token_context_cuda_graph.py::test_zero_runtime_draft_speculation_after_final_token_reuse[v2]": 26.2569375, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_batch_generation": 18.130119363395224, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_cold_start": 21.948513368983956, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_field_completeness": 12.267211796246649, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_full_block_reuse": 19.860912, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_long_context": 29.62158510638298, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_partial_block_reuse": 25.294635135135138, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_rapid_fire": 13.005842391304348, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_shared_prefix": 19.404803278688522, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens": 291.17222857142855, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens": 126.47582857142858, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_eviction": 31.50208653846154, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_batch_size_limited": 130.98199723756906, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_block_reuse": 162.45401400560226, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill": 181.319, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_eviction_block_reuse": 20.081730245231608, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_multi_request": 157.2019005524862, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_with_eviction": 20.046337950138504, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[cuda_graph]": 17.19818487394958, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[no_cuda_graph]": 28.09476404494382, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_overlap": 17.407029810298102, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_with_block_reuse": 43.698, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[non_overlap]": 135.2105218579235, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[overlap]": 154.92498382749326, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_partial_block_reuse": 170.93294623655913, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_token_budget_limited": 133.1101049723757, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_v2_vs_v1_basic": 135.72120055710306, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_chunked_prefill": 133.00545658263306, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_eviction": 46.97228690807799, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_multi_adapter_v2": 164.67938292011019, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_v2": 190.69342222222224, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[guaranteed-chunked]": 159.4352530120482, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[max-util-chunked]": 159.47862611275963, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[no-overlap-chunked]": 169.5043982300885, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-chunked]": 162.48168338557994, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-no-chunked]": 162.01903592814372, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[python-scheduler]": 168.00891788856305, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[swa-chunked]": 133.72491875, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix_smoke": 132.2567359050445, - "llmapi/test_llm_api_connector.py::test_connector_async_onboard[False]": 17.844191326530613, - "llmapi/test_llm_api_connector.py::test_connector_async_onboard[True]": 17.94226785714286, - "llmapi/test_llm_api_connector.py::test_connector_async_save[False]": 17.79670512820513, - "llmapi/test_llm_api_connector.py::test_connector_async_save[True]": 12.164752551020408, - "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[False]": 23.325279792746116, - "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[True]": 23.173549872122763, - "llmapi/test_llm_api_connector.py::test_connector_e2e_persistent_cache": 21.657964194373402, - "llmapi/test_llm_api_connector.py::test_connector_multi_request": 9.827329048843188, - "llmapi/test_llm_api_connector.py::test_connector_priorities": 11.924429667519181, - "llmapi/test_llm_api_connector.py::test_connector_priorities_default": 11.87047193877551, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[attention_dp]": 3.934010101010101, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[beam_search]": 9.53984693877551, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[host_offloading]": 11.150828205128207, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[False]": 14.05459383033419, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[True]": 13.803569974554708, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[False]": 11.044869897959185, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[True]": 10.860375, - "llmapi/test_llm_api_connector.py::test_connector_simple[False]": 11.83360358056266, - "llmapi/test_llm_api_connector.py::test_connector_simple[True]": 11.803266666666666, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-bart-large-cnn]": 31.893891414141414, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-bart-large-cnn]": 15.499701803051318, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-bart-large-cnn]": 22.250857142857143, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 28.036251908396945, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-bart-large-cnn]": 6.71089847715736, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-bart-large-cnn]": 13.861859174964438, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-bart-large-cnn]": 22.041833333333333, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 19.904857142857143, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 7.540977157360405, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 15.9137, - "llmapi/test_llm_api_pytorch_bart.py::test_mbart_pytorch_generate_encoder_decoder_end_to_end": 18.597017811704834, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[cudagraph]": 83.4373853211009, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[eager]": 81.5805, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-t5-small]": 5.8821772151898735, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 11.94230964467005, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-xl]": 36.057, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-t5-small]": 18.94985714285714, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-base]": 19.276714285714288, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-small]": 12.308420903954802, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-t5-small]": 11.595129760225669, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-t5-small]": 5.213745547073791, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-tp2-t5-small]": 21.989346534653464, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-off-greedy-byt5-small]": 18.854222222222223, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.96542857142857, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-t5-small]": 18.41714285714286, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-t5-small]": 5.065276649746193, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 18.676571428571428, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-t5-small]": 18.371142857142857, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.455428571428573, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-t5-small]": 17.964, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 17.551571428571428, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-t5-small]": 18.958428571428573, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.94242857142857, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-t5-small]": 18.021428571428572, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_context_generation_batch": 5.631847715736041, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-flan-t5-small]": 15.022181818181817, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-t5-small]": 15.552, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-t5-small]": 5.275149367088607, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-t5-small]": 16.256125, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_beam_search[bf16-kv-v1-decoder-graphs-on-beam2]": 9.641889502762432, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_beam_search[fp32-kv-v1-graphs-off-beam2]": 8.5922, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[bf16-kv-v1-decoder-graphs-on-greedy]": 12.066636363636364, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[bf16-kv-v2-decoder-graphs-on-greedy]": 19.83623324396783, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp16-kv-v1-graphs-off-greedy]": 10.803600000000001, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v1-graphs-off-greedy-tp2]": 25.9721875, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v1-graphs-requested-greedy]": 5.953636363636364, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v2-graphs-off-greedy]": 6.5942, - "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_transcribe_end_to_end": 14.981148247978437, - "llmapi/test_llm_examples.py::test_llmapi_chat_example": 67.295, - "llmapi/test_llm_examples.py::test_llmapi_example_distributed_tp2": 53.464285714285715, - "llmapi/test_llm_examples.py::test_llmapi_example_guided_decoding": 59.278842083675144, - "llmapi/test_llm_examples.py::test_llmapi_example_inference": 66.28026499589154, - "llmapi/test_llm_examples.py::test_llmapi_example_inference_async": 46.420170701682395, - "llmapi/test_llm_examples.py::test_llmapi_example_inference_async_streaming": 52.58203203285421, - "llmapi/test_llm_examples.py::test_llmapi_example_logits_processor": 45.783905916187344, - "llmapi/test_llm_examples.py::test_llmapi_example_multilora": 59.31795570139459, - "llmapi/test_llm_examples.py::test_llmapi_kv_cache_connector[Qwen2-0.5B]": 88.863, - "llmapi/test_llm_examples.py::test_llmapi_quickstart": 45.94649178307313, - "llmapi/test_llm_examples.py::test_llmapi_quickstart_atexit": 277.5743333333333, - "llmapi/test_llm_examples.py::test_llmapi_server_example": 155.23228571428572, - "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_mtp": 218.2092435897436, - "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_ngram": 92.54680097087378, - "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_context": 0.9749723756906077, - "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_generation": 0.919544474393531, - "perf/host_perf/test_module_sampler.py::test_sampler_update_greedy[greedy_bs8]": 1.1856621253405994, - "perf/host_perf/test_module_sampler.py::test_sampler_update_stop_words[stopwords_bs32]": 1.2631570247933885, - "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_gen_only_bs8]": 0.8479316939890711, - "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_mixed_32gen_4ctx]": 0.9042729729729729, - "perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:8192]": 243.04695381062356, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 1999.384, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 408.2405, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 960.0195714285715, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 379.282625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2356.765423255814, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 456.45875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 2758.6535, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 387.9955, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 2762.70475, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 198.09625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 572.877202764977, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1376.504375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2174.338375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 512.32025, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2131.6131612903223, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 316.531625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 407.063537037037, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 4896.811888888889, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2131.481125, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1495.524375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 2059.856375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1268.812375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 2882.679875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1273.481375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1972.6055, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 466.771375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 1304.3386, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_tep8_mtp3]": 541.446, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 651.103, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 398.36628571428577, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_dep4_mtp1_8k1k]": 905.3361290322581, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_1k8k]": 950.800625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_8k1k]": 561.985, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_1k8k]": 601.254875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 402.47142396313365, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_ad_blackwell-r1_fp8_ad_ws8_1k1k]": 715.4331428571428, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_blackwell-r1_fp8_tp8_6k1k]": 1442.0042857142857, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_dep8_mtp1_8k1k]": 858.155875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_tep8_mtp3_8k1k]": 380.380875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_dep4_mtp1_8k1k]": 917.161625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_tep4_mtp3_8k1k]": 489.380875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_deepseek_v32_fp4_2_nodes_grace_blackwell-dsv32_fp4_dep8_trtllm_lpc_mnnvl_8k1k]": 453.6847142857143, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_adp_2k1k]": 590.84, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_235b_a22b_fp8_hopper-qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1k]": 946.8511428571428, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_32b_fp8_hopper-qwen3_32b_fp8_tp2_6k1k]": 768.5237142857143, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gemma4_26b_a4b_nvfp4_blackwell-gemma4_26b_a4b_nvfp4_tp1_1k1k]": 322.5816615384615, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 578.0965, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_tep8_mtp3_8k1k]": 574.7941428571428, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 688.625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 595.070625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_dep4_1k8k]": 1280.064875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tep2_1k8k]": 802.94175, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 589.688, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp2_1k8k]": 501.079625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_deepseek_v3_lite-v3lite_fp8_bs8_128_256]": 569.2608436657681, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b-llama8b_fp16_bs8_128_256]": 264.4596918767507, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b_spec_decode-llama8b_spec_bs1_128_128]": 177.14550555555553, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_2_nodes_grace_blackwell-k25_thinking_fp4_dep8_32k8k]": 1465.1101666666668, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_32k8k]": 1438.051875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_8k1k]": 764.6005, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_32k8k]": 562.570625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_8k1k]": 340.288625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_dep4_8k1k]": 837.4493411214953, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_tep4_8k1k]": 447.6755, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_1000_1000]": 462.81728571428573, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_512_32]": 250.122875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]": 708.9645, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]": 599.9765, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep4_mtp3_8k1k]": 567.198875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tp4_8k1k]": 518.17875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_ad_blackwell-super_ad_ws4_1k1k]": 579.10435, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_mtp_ad_blackwell-super_mtp_ad_ws4_1k1k]": 541.9476666666666, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 4236.98, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2463.5651428571427, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 3260.8825714285713, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 3234.713142857143, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 865.6539590909091, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2143.3998571428574, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2017.474, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 6895.875, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2307.246111111111, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1396.0565, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 2622.707125, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1470.1127142857142, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1339.9375714285713, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 3237.387563218391, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1942.96, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 871.0367857142857, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 508.911, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 554.807, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2544.6345714285712, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 862.036375, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 539.3786, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 427.9478, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 885.85225, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 635.96075, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 871.6156, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 579.8865, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 1007.6004, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 322.9055, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con128_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 298.158, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 279.817375, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 865.159, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 497.62114285714284, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 313.13625, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 331.31875, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 2963.189, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1329.4505, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 1577.9463333333333, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1356.0965, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1260.772, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 1073.543, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 605.6804285714285, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 598.488, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 878.816, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 469.214023923445, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-flux2_blackwell-flux2_fp8_cfg1_ulysses4_teacache_on]": 333.64025, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_i2v_cfg2_ulysses4_compile_on]": 536.826875, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_t2v_cfg2_ulysses4_compile_on]": 476.725875, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_nvfp4_i2v_cfg2_ulysses4_compile_on]": 472.473875, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan21_t2v_14b_blackwell-wan21_14b_nvfp4_trtllm_cfg2_ulysses4_teacache_on]": 511.6845, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan22_i2v_a14b_blackwell-wan22_i2v_a14b_nvfp4_trtllm_cfg2_ulysses4]": 458.00059999999996, - "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp1_4instances]": 103.69261616161616, - "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp2_2instances]": 107.46637373737374, - "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-GUARANTEED_NO_EVICT-pytorch-stress-test]": 820.9262857142857, - "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-MAX_UTILIZATION-pytorch-stress-test]": 736.4245714285714, - "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 63.3935, - "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3_depth_1_tree[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 59.22325, - "test_e2e.py::test_get_ci_container_port": 1.294469543147208, - "test_e2e.py::test_openai_chat_example[pytorch]": 568.4271944444445, - "test_e2e.py::test_openai_chat_guided_decoding[meta-llama/Llama-3.1-8B-Instruct]": 72.82539402985076, - "test_e2e.py::test_openai_chat_guided_decoding[openai/gpt-oss-120b]": 170.5241832884097, - "test_e2e.py::test_openai_chat_harmony": 180.47430188679243, - "test_e2e.py::test_openai_chat_harmony_perf_metrics": 168.00521202531647, - "test_e2e.py::test_openai_chat_multimodal_example": 128.08525063938617, - "test_e2e.py::test_openai_chat_with_logit_bias[torch_sampler]": 85.82997073170732, - "test_e2e.py::test_openai_chat_with_logit_bias[trtllm_sampler]": 85.85354501216544, - "test_e2e.py::test_openai_completions_example[pytorch]": 255.78602307692307, - "test_e2e.py::test_openai_completions_with_logit_bias[torch_sampler]": 83.29217518248174, - "test_e2e.py::test_openai_completions_with_logit_bias[trtllm_sampler]": 83.41751707317073, - "test_e2e.py::test_openai_health": 69.34942857142856, - "test_e2e.py::test_openai_kv_cache_contamination": 1558.114663265306, - "test_e2e.py::test_openai_lora": 184.53052987012987, - "test_e2e.py::test_openai_misc_example[pytorch]": 338.3785968992248, - "test_e2e.py::test_openai_mmencoder_example": 132.4975871794872, - "test_e2e.py::test_openai_perf_metrics": 83.57704303797469, - "test_e2e.py::test_openai_post_processor": 367.14763144329896, - "test_e2e.py::test_openai_prometheus": 91.01292583120204, - "test_e2e.py::test_openai_reasoning[pytorch]": 253.63865025906736, - "test_e2e.py::test_openai_responses": 318.4291543026706, - "test_e2e.py::test_openai_responses_entrypoint": 180.99946272493574, - "test_e2e.py::test_openai_tool_call": 92.28577581863979, - "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-120B-gpt_oss/gpt-oss-120b]": 405.7005, - "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-20B-gpt_oss/gpt-oss-20b]": 74.43636476426799, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B]": 78.70912158808933, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]": 60.669947300771206, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta-Llama-3.1-8B]": 80.62433642384106, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]": 156.90823631840797, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_fp8_hf-Qwen3/saved_models_Qwen3-30B-A3B_fp8_hf]": 116.35483084577113, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_nvfp4_hf-Qwen3/saved_models_Qwen3-30B-A3B_nvfp4_hf]": 94.89845024875622, - "test_e2e.py::test_ptp_quickstart_advanced_bs1": 97.3342, - "test_e2e.py::test_ptp_quickstart_advanced_deepseek_v3_lite_4gpus_adp_balance[DeepSeek-V3-Lite-FP8-DeepSeek-V3-Lite/fp8]": 95.85601020408163, - "test_e2e.py::test_ptp_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 75.93503878116343, - "test_e2e.py::test_ptp_quickstart_advanced_mixed_precision": 90.59626630434782, - "test_e2e.py::test_ptp_quickstart_advanced_mtp[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 168.06148467966574, - "test_e2e.py::test_ptp_quickstart_advanced_mtp_eagle[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 250.64657452574528, - "test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct]": 70.2027741046832, - "test_e2e.py::test_ptp_quickstart_bert[TRTLLM-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 20.47236647955092, - "test_e2e.py::test_ptp_quickstart_bert[VANILLA-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 6.805838813151564, - "test_e2e.py::test_trtllm_bench_help_sanity[meta-llama/Llama-3.1-8B]": 46.12830721003135, - "test_e2e.py::test_trtllm_bench_invalid_token_pytorch[TinyLlama-1.1B-Chat-v1.0-TinyLlama-1.1B-Chat-v1.0]": 77.25574680306904, - "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-non-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.06281155015198, - "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 88.87241614906831, - "test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b]": 123.27977142857144, - "test_e2e.py::test_trtllm_bench_mgmn": 113.6422857142857, - "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-False-False]": 117.62252560386473, - "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-instruct-hf-fp8-True-True]": 67.63684076433121, - "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-]": 115.25042385786803, - "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-enable_request_rate]": 32.17077581120944, - "test_e2e.py::test_trtllm_serve_example": 135.66885714285712, - "test_e2e.py::test_trtllm_serve_lora_example": 93.26694545454545, - "test_e2e.py::test_trtllm_serve_multimodal_example": 176.18389285714287, - "test_e2e.py::test_trtllm_serve_top_logprobs[pytorch]": 85.50150381679389, - "test_fmha.py::test_fmha": 2383.811928571429, - "thirdparty/test_cmake_third_party.py::test_cmake_listfiles": 1.309677749360614, - "thirdparty/test_git_modules.py::test_gitmodules": 1.2179113924050633, - "unittest/_torch/attention": 1427.1057219730942, - "unittest/_torch/attention --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1447.3447479452057, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_kernel.py": 26.8552816091954, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_module.py": 132.06273595505618, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_tf32.py": 23.715, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_cache_manager.py": 54.658080332409966, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_indices_transform.py": 58.47613623978202, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_o_proj.py": 37.03518105849582, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_sparse_mla.py": 55.00545303867403, - "unittest/_torch/attention/sparse/dsa/test_dsa_fp4_indexer.py": 23.768575757575757, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_cute_dsl_fp8_indexer_q_gemm_rope_fp4_matches_unfused": 47.0700824742268, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_dsv32_prefill_shape": 24.42627, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_matches_deepgemm": 20.244656565656566, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_noncontiguous_split": 2.1675763440860214, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_shape_dispatch": 22.984191919191918, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_indexer_k_cache_gather_contiguous_fp4": 20.064323232323233, - "unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py": 725.6988395721925, - "unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py": 344.4336289473684, - "unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1976.4200219178083, - "unittest/_torch/attention/sparse/test_sparse_mla_forward.py": 122.47299180327869, - "unittest/_torch/attention/test_attention_mla.py": 133.47027261904762, - "unittest/_torch/auto_deploy": 25.987419558359623, - "unittest/_torch/compilation": 26.45571775417299, - "unittest/_torch/custom_ops/test_deepseek_v4_q_norm.py": 23.911654494382024, - "unittest/_torch/debugger": 21.257522302158275, - "unittest/_torch/distributed": 20.148352380952378, - "unittest/_torch/executor": 312.95409486780716, - "unittest/_torch/executor/test_async_transfer_manager.py": 26.81270284237726, - "unittest/_torch/executor/test_disagg_index_mapper_early_release.py": 27.300713917525773, - "unittest/_torch/executor/test_error_classification.py": 25.742147582697203, - "unittest/_torch/executor/test_hang_detector_kill.py": 41.70411578947368, - "unittest/_torch/executor/test_kv_cache_budget_split.py": 26.940844387755103, - "unittest/_torch/executor/test_kv_cache_compression_manager.py": 27.190636597938145, - "unittest/_torch/executor/test_kv_cache_estimation.py": 27.18597461928934, - "unittest/_torch/executor/test_kv_cache_v2_capacity_only.py": 25.351969072164948, - "unittest/_torch/executor/test_kv_pool_rebalance.py": 27.080116279069767, - "unittest/_torch/executor/test_model_loader_mx.py": 23.324768304914745, - "unittest/_torch/executor/test_overlap_scheduler.py": 762.4326900584796, - "unittest/_torch/executor/test_scheduler_serializable_output.py": 25.51113810741688, - "unittest/_torch/flashinfer/test_trtllm_flashinfer_symbol_collision.py": 1010.8754930362117, - "unittest/_torch/lora": 21.032186781609195, - "unittest/_torch/memory": 20.313926751592355, - "unittest/_torch/misc": 207.61796048632218, - "unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy": 42.2076, - "unittest/_torch/modeling -k \"modeling_gemma3\"": 246.71202580645163, - "unittest/_torch/modeling -k \"modeling_gpt_oss\"": 98.51157242339833, - "unittest/_torch/modeling -k \"modeling_llama\"": 134.55647529069768, - "unittest/_torch/modeling -k \"modeling_mixtral\"": 77.18025104022192, - "unittest/_torch/modeling -k \"modeling_nemotron_nano_v2_vl\"": 457.9309539641944, - "unittest/_torch/modeling -k \"modeling_nemotron_nas\"": 39.4559902676399, - "unittest/_torch/modeling -k \"modeling_out_of_tree\"": 143.6954725014784, - "unittest/_torch/modeling -k \"modeling_phi3\"": 33.97074209245742, - "unittest/_torch/modeling -k \"modeling_qwen\"": 701.1010346820809, - "unittest/_torch/modeling -k \"modeling_qwen_moe\"": 298.2582273838631, - "unittest/_torch/modeling -k \"modeling_siglip\"": 190.55107537688443, - "unittest/_torch/modeling -k \"modeling_speculative\"": 27.12980684596577, - "unittest/_torch/modeling -k \"modeling_vila\"": 42.61871392405063, - "unittest/_torch/modeling -k \"modeling_whisper\"": 22.399005917159766, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_multimodal_26b_dummy": 174.59721359223303, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_26b_dummy": 176.25729368575625, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_31b_dummy": 207.95196536796536, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy": 178.88090070921987, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[256]": 182.0266661991585, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[512]": 207.8059829302987, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e4b_dummy": 188.1576099585062, - "unittest/_torch/modeling/test_gemma4_multimodal.py": 26.365072625698325, - "unittest/_torch/modeling/test_modeling_afmoe.py": 70.01261643835618, - "unittest/_torch/modeling/test_modeling_cohere2.py": 77.30710997442456, - "unittest/_torch/modeling/test_modeling_cosmos3.py::TestCosmos3::test_all": 25.3536120906801, - "unittest/_torch/modeling/test_modeling_deepseekv4.py": 57.31769696969697, - "unittest/_torch/modeling/test_modeling_exaone_moe.py": 252.087638483965, - "unittest/_torch/modeling/test_modeling_gemma4.py": 80.82831623931624, - "unittest/_torch/modeling/test_modeling_gemma4_unified.py": 20.30774293785311, - "unittest/_torch/modeling/test_modeling_llava_next.py::TestLlavaNext::test_all": 35.675928030303034, - "unittest/_torch/modeling/test_modeling_llava_next.py::test_llava_next_expand_prompt_token_ids_for_mm": 14.221383165829145, - "unittest/_torch/modeling/test_modeling_minicpmv4_6.py": 28.253985454545454, - "unittest/_torch/modeling/test_modeling_mistral.py": 40.019712820512815, - "unittest/_torch/modeling/test_modeling_nemotron_h.py::test_nemotron_h_sanity": 181.17747040498443, - "unittest/_torch/modeling/test_modeling_parakeet.py": 43.805466836734695, - "unittest/_torch/modeling/test_modeling_pixtral.py": 35.962048969072164, - "unittest/_torch/modeling/test_modeling_pixtral.py::test_tensor_parallelism": 42.47271568627451, - "unittest/_torch/modeling/test_modeling_qwen2_5vl.py::TestQwen2_5_VL::test_all": 76.01826083188908, - "unittest/_torch/modeling/test_modeling_qwen3_5_vl.py::TestQwen3_5VL::test_all": 71.38927666666667, - "unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py::TestQwen3_5MoeVL::test_all": 79.15871782945737, - "unittest/_torch/modeling/test_modeling_qwen3vl.py::TestQwen3VL::test_all": 96.50474396551724, - "unittest/_torch/modeling/test_modeling_qwen3vl.py::test_qwen3vl_init_preserves_caller_quant_config": 11.537473136915077, - "unittest/_torch/modeling/test_modeling_qwen3vl_moe.py::TestQwen3VLMoe::test_all": 82.38036465517241, - "unittest/_torch/modeling/test_modeling_radio.py": 29.20512853470437, - "unittest/_torch/modeling/test_modeling_step3p7.py": 28.51029411764706, - "unittest/_torch/modeling/test_modeling_step3p7vl.py": 40.39187341772151, - "unittest/_torch/modeling/test_multimodal_encoder_graph.py": 20.74985975609756, - "unittest/_torch/modeling/test_multimodal_encoder_mixin.py": 25.43182005141388, - "unittest/_torch/modeling/test_nemotron_nano_preprocessing.py": 63.15041176470588, - "unittest/_torch/models/checkpoints": 63.97590512820513, - "unittest/_torch/models/test_qwen3_next_moe_quant.py": 25.30277272727273, - "unittest/_torch/modules/dwdp/test_dwdp_fixup_moe_backends.py": 25.54453746770026, - "unittest/_torch/modules/dwdp/test_dwdp_manager.py": 25.44528165374677, - "unittest/_torch/modules/dwdp/test_dwdp_mapping.py": 25.49276546391753, - "unittest/_torch/modules/dwdp/test_dwdp_peer_ranges.py": 25.482807291666667, - "unittest/_torch/modules/fused_moe/test_deepgemm_fused_expand_quant.py": 22.72942074927954, - "unittest/_torch/modules/fused_moe/test_deepgemm_fused_gather_finalize.py": 23.010556603773583, - "unittest/_torch/modules/fused_ops/test_fused_qk_norm_rope_gate.py": 24.518203296703298, - "unittest/_torch/modules/fused_ops/test_gelu_tanh_mul_fp4_quant.py": 25.7476404494382, - "unittest/_torch/modules/fused_ops/test_rmsnorm_fp4_quant.py": 24.488736263736264, - "unittest/_torch/modules/fused_ops/test_rmsnorm_residual_add.py": 24.566434065934068, - "unittest/_torch/modules/mamba": 743.5672699507388, - "unittest/_torch/modules/moe/test_cute_dsl_b12x_moe_backend.py": 21.25159801488834, - "unittest/_torch/modules/moe/test_megamoe_streaming_load.py": 21.72980769230769, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 95.0835, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL\"": 261.44885714285715, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTLASS\"": 1145.8342608695652, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 45.4305, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM\"": 62.880714285714284, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DENSEGEMM\"": 664.1701428571429, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_CUTEDSL\"": 121.23383333333332, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_DEEPGEMM\"": 28.824571428571428, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"TRTLLM\"": 121.97892307692308, - "unittest/_torch/modules/moe/test_moe_backend.py::test_trtllm_bf16_unquantized_moe": 934.3070607344632, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm": 260.6249773755656, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_boundary": 227.2692, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_postquant": 123.30966502463053, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_nccl_ep_cuda_graph_replay_uses_updated_routing": 45.36088387096775, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTEDSL\"": 96.90683838383839, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES and not W4A8 and not MXFP8\"": 79.44716666666667, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES\"": 71.319, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8_BLOCK_SCALES\"": 60.91212, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and MXFP8 and not W4A8\"": 74.81355555555557, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and NVFP4\"": 72.06288, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A16_MXFP4\"": 70.43882653061225, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_AWQ\"": 62.54622, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_FP8\"": 80.16814141414142, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_MXFP8\"": 107.48528282828282, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W8A16\"": 78.97271573604061, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 79.17475757575757, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MARLIN\"": 81.26303157894736, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_CUTEDSL\"": 62.499848484848485, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_DEEPGEMM\"": 54.97358585858586, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and FP8_BLOCK_SCALES\"": 61.35357575757576, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and NVFP4 and not W4A8\"": 58.35683838383838, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A16_MXFP4\"": 66.28399, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_MXFP4_MXFP8\"": 52.592858585858586, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_NVFP4_FP8\"": 151.46551515151515, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 63.586188118811876, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 47.33725742574257, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 56.93824752475248, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 80.86554455445544, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.643316831683165, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_CUTEDSL-quant=NVFP4-routing=DeepSeekV3]": 58.67727722772277, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_DEEPGEMM-quant=W4A8_MXFP4_MXFP8-routing=DeepSeekV3]": 55.511158415841585, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 26.81741584158416, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 26.794633663366337, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 26.845336633663365, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 48.804900990099014, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.747702970297034, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.98811881188119, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb": 48.077839999999995, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_CUTEDSL\"": 57.46130693069307, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_DEEPGEMM\"": 52.91237623762377, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"not None\"": 25.53242, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 47.52238613861386, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.06145544554455, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 48.202405940594055, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 89.27855492957747, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and None\"": 27.112461538461538, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and not None\"": 149.8221215805471, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 50.51278611111111, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DENSEGEMM\"": 454.09614, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_CUTEDSL\"": 114.45545983379502, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_DEEPGEMM\"": 42.27300540540541, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 65.58272594752187, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 32.970935754189945, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 31.079560344827588, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 29.457871148459382, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 30.559139130434783, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.20927576601671, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.70960465116279, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 35.24991267605634, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 28.97459824046921, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 58.56761344537815, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.696147058823527, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.float16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.576210407239817, - "unittest/_torch/modules/test_awq_quantization.py": 20.355017509727627, - "unittest/_torch/modules/test_engram.py": 95.12348051948052, - "unittest/_torch/modules/test_fp4_num_tokens_slice.py": 21.562614084507043, - "unittest/_torch/modules/test_fused_activation_quant.py": 21.362952153110047, - "unittest/_torch/modules/test_fused_add_rms_norm_quant.py": 22.13294797687861, - "unittest/_torch/modules/test_fused_rmsnorm_fp4_quantize.py": 28.106347578347577, - "unittest/_torch/modules/test_gemma4_fused_qkv_prep.py": 25.284024861878454, - "unittest/_torch/modules/test_group_rmn_norm.py": 26.97853846153846, - "unittest/_torch/modules/test_mhc.py": 30.238890876565296, - "unittest/_torch/modules/test_mla_helix.py": 41.2383063583815, - "unittest/_torch/modules/test_moe_host_sharer.py": 22.98750459921156, - "unittest/_torch/modules/test_moe_load_balancer.py": 22.217951550387596, - "unittest/_torch/modules/test_moe_routing.py": 46.00725428571428, - "unittest/_torch/modules/test_rotary_embedding.py": 21.841127167630056, - "unittest/_torch/modules/test_triton_linear.py": 21.558213051823415, - "unittest/_torch/modules/tests_lora_modules": 366.0990338983051, - "unittest/_torch/multi_gpu -m \"not post_merge\"": 274.7847525773196, - "unittest/_torch/multi_gpu_modeling -k \"deepseek\"": 332.8804090909091, - "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp1-bf16-trtllm-deepseekv3_lite]": 83.28676470588235, - "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp4-bf16-trtllm-deepseekv3_lite]": 114.40690045248869, - "unittest/_torch/multimodal -k \"not nemotron_nano_v2_vl_fp8\"": 764.3596734693878, - "unittest/_torch/multimodal/test_mm_encoder_standalone.py -k \"qwen3_30b_a3b_fp8\"": 327.08215211267606, - "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu2\"": 223.64610000000002, - "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu4\"": 458.21805050505054, - "unittest/_torch/ray_orchestrator/multi_gpu/test_inflight_weight_update.py": 84.21391, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part2\"": 624.9352843137256, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part3\"": 579.3408627450981, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part5\"": 577.8696666666666, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py::test_llm_partial_update_weights_qwen35_35b_bf16_tp8_cuda_graph": 210.214125, - "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp1_4instances]": 268.0139898989899, - "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp2_2instances]": 282.0997121212121, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part0\"": 274.29658, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part1\"": 273.73165, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part2\"": 147.88863, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part3\"": 148.11237, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part4\"": 204.57628, - "unittest/_torch/ray_orchestrator/single_gpu/test_cache_transceiver_comm.py": 48.72692219020173, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_sleep.py": 112.46935964912281, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part0\"": 711.4720964912281, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part1\"": 661.9107280701754, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part2\"": 486.40483040935675, - "unittest/_torch/sampler -k \"not test_speculative_d2h_parity_real_predictor\"": 631.3018140625001, - "unittest/_torch/sampler/test_beam_search.py": 444.93877317073174, - "unittest/_torch/sampler/test_beam_search_speculative_d2h.py": 127.7736968215159, - "unittest/_torch/sampler/test_logits_logprobs.py": 260.9117658536585, - "unittest/_torch/sampler/test_penalties.py": 56.53724752475248, - "unittest/_torch/sampler/test_token_ban.py": 26.095191964285714, - "unittest/_torch/sampler/test_torch_sampler.py": 226.65034025974023, - "unittest/_torch/sampler/test_trtllm_sampler.py": 92.73561953727507, - "unittest/_torch/speculative/hw_agnostic": 638.9258896551723, - "unittest/_torch/speculative/test_capture_override_leak.py": 20.019795454545456, - "unittest/_torch/speculative/test_eagle3.py": 698.1639762611276, - "unittest/_torch/speculative/test_rejection_buffers_guard.py": 20.62444705882353, - "unittest/_torch/speculative/test_sa_hybrid_state_promotion.py": 20.992551724137932, - "unittest/_torch/test_model_config.py": 25.45066919191919, - "unittest/_torch/test_tensor_lru_cache.py": 25.701808510638298, - "unittest/_torch/test_torch_multi_arange.py": 44.52563636363636, - "unittest/_torch/thop/parallel": 2201.6000647103083, - "unittest/_torch/thop/parallel/test_finegrained_mixed_dtype_gemm.py": 27.07066501240695, - "unittest/_torch/thop/parallel/test_w4a16_linear.py": 21.95623076923077, - "unittest/_torch/thop/parallel/test_w4a8_linear.py": 20.894305210918116, - "unittest/_torch/thop/parallel/test_weight_only_quant_gemm.py": 60.234856079404466, - "unittest/_torch/thop/parallel/test_weight_only_quant_linear.py": 27.418575682382137, - "unittest/_torch/thop/parallel_hw_agnostic": 467.797237012987, - "unittest/_torch/thop/serial": 53.24263645224171, - "unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py": 328.0071551724138, - "unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py": 210.79964705882352, - "unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py": 233.11, - "unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py": 202.95143220338983, - "unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py": 144.93126495726497, - "unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py": 70.44280672268907, - "unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py": 189.4298151260504, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py": 44.64089830508475, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py": 85.23357264957265, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py": 64.13015254237288, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py": 254.18874789915967, - "unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py": 194.66533898305084, - "unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py": 300.4546120689655, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py": 65.64371186440678, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py": 371.2901440677966, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py": 128.87157627118643, - "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py": 214.06044347826085, - "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_multinode.py": 22.233025423728815, - "unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py": 76.19292372881357, - "unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py": 261.1740084745763, - "unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py": 206.67422807017545, - "unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py": 521.5535932203389, - "unittest/_torch/visual_gen/test_attention_cute_dsl.py": 161.75197866666667, - "unittest/_torch/visual_gen/test_attention_cute_dsl_vsa.py": 43.49738057742782, - "unittest/_torch/visual_gen/test_attention_integration.py": 107.21697574123989, - "unittest/_torch/visual_gen/test_attention_perf.py": 219.80059466666668, - "unittest/_torch/visual_gen/test_attention_trtllm_sage.py": 204.94523116883119, - "unittest/_torch/visual_gen/test_cache_dit.py": 170.39281714285715, - "unittest/_torch/visual_gen/test_cosmos3_distilled.py": 36.37447619047619, - "unittest/_torch/visual_gen/test_cosmos3_pipeline.py": 152.43147802197802, - "unittest/_torch/visual_gen/test_cosmos3_transformer.py": 94.60373802816902, - "unittest/_torch/visual_gen/test_flux2_image_conditioning.py": 19.8582, - "unittest/_torch/visual_gen/test_flux_attention.py": 27.198191489361704, - "unittest/_torch/visual_gen/test_flux_infer.py": 25.557968523002423, - "unittest/_torch/visual_gen/test_flux_pipeline.py": 489.605927027027, - "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxCombinedOptimizations::test_all_optimizations_combined": 76.62864705882352, - "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxParallelism::test_ulysses_2gpu_correctness": 64.04692796610169, - "unittest/_torch/visual_gen/test_flux_transformer.py": 45.653248677248676, - "unittest/_torch/visual_gen/test_ltx2_attention.py": 24.870206434316355, - "unittest/_torch/visual_gen/test_ltx2_pipeline.py": 727.259218579235, - "unittest/_torch/visual_gen/test_ltx2_transformer.py": 27.478767441860466, - "unittest/_torch/visual_gen/test_model_loader.py": 44.97012464589235, - "unittest/_torch/visual_gen/test_profiler.py": 26.178076923076922, - "unittest/_torch/visual_gen/test_quant_ops.py": 22.4151689373297, - "unittest/_torch/visual_gen/test_qwen_image_infer.py": 25.41444711538462, - "unittest/_torch/visual_gen/test_qwen_image_pipeline.py": 25.53377886977887, - "unittest/_torch/visual_gen/test_teacache.py": 20.578375, - "unittest/_torch/visual_gen/test_tensor_payload.py": 25.7702, - "unittest/_torch/visual_gen/test_trtllm_serve_e2e.py": 493.3113262032085, - "unittest/_torch/visual_gen/test_trtllm_serve_endpoints.py": 29.387322493224932, - "unittest/_torch/visual_gen/test_visual_gen_args.py": 22.076759894459105, - "unittest/_torch/visual_gen/test_visual_gen_params.py": 26.502983565107456, - "unittest/_torch/visual_gen/test_visual_gen_utils.py": 21.9845, - "unittest/_torch/visual_gen/test_wan21_i2v_pipeline.py": 201.90125721784779, - "unittest/_torch/visual_gen/test_wan21_i2v_teacache.py": 332.5180740740741, - "unittest/_torch/visual_gen/test_wan21_t2v_pipeline.py": 183.95970684931507, - "unittest/_torch/visual_gen/test_wan21_t2v_teacache.py": 232.78348850574713, - "unittest/_torch/visual_gen/test_wan21_t2v_teacache_user_coefficients.py": 52.60576533333334, - "unittest/_torch/visual_gen/test_wan22_i2v_pipeline.py": 270.040718579235, - "unittest/_torch/visual_gen/test_wan22_i2v_teacache.py": 138.1834361702128, - "unittest/_torch/visual_gen/test_wan22_t2v_pipeline.py": 256.53519293478263, - "unittest/_torch/visual_gen/test_wan22_t2v_teacache.py": 130.60224316939892, - "unittest/_torch/visual_gen/test_wan22_ti2v_5b_pipeline.py": 123.94177427821522, - "unittest/_torch/visual_gen/test_wan_transformer.py": 77.91112391930837, - "unittest/_torch/visual_gen/test_wan_vae.py": 67.90559340659341, - "unittest/_torch/visual_gen/test_wan_vsa_pipeline.py": 119.16156484149856, - "unittest/_torch/visual_gen/test_warmup.py": 21.650458666666665, - "unittest/_torch/weight_sharing": 25.79753886010363, - "unittest/api_stability": 30.174709183673468, - "unittest/auto_deploy/multigpu/compile": 43.57656435643565, - "unittest/auto_deploy/multigpu/smoke": 154.60278606965176, - "unittest/auto_deploy/multigpu/transformations": 2054.764138613861, - "unittest/auto_deploy/multigpu/transformations/library/test_allreduce_residual_rmsnorm_fusion.py -k \"strategy_auto\"": 121.42506, - "unittest/auto_deploy/multigpu/transformations/library/test_tp_sharding.py::test_moe_tp_shard_nvfp4": 71.84925, - "unittest/auto_deploy/singlegpu/compile": 70.71696694214876, - "unittest/auto_deploy/singlegpu/custom_ops": 525.2752493074793, - "unittest/auto_deploy/singlegpu/custom_ops/attention/test_triton_attention.py::TestSDPADispatch": 39.1398, - "unittest/auto_deploy/singlegpu/custom_ops/mamba/test_flashinfer_mamba_cached_op.py": 48.05372479564033, - "unittest/auto_deploy/singlegpu/custom_ops/moe/test_ad_moe_op.py": 29.51027642276423, - "unittest/auto_deploy/singlegpu/custom_ops/moe/test_trtllm_moe.py": 41.36786178861789, - "unittest/auto_deploy/singlegpu/custom_ops/quantization/test_quant.py": 50.740886178861786, - "unittest/auto_deploy/singlegpu/mlir": 29.758081267217634, - "unittest/auto_deploy/singlegpu/models": 233.6499502074689, - "unittest/auto_deploy/singlegpu/shim": 96.55770539419086, - "unittest/auto_deploy/singlegpu/smoke": 1689.3055637119114, - "unittest/auto_deploy/singlegpu/smoke/test_ad_build_small_single.py -k \"Nemotron-3-Nano-30B-A3B-FP8 or Nemotron-Nano-3-30B-A3.5B-dev or Llama-4-Scout\"": 249.0911842818428, - "unittest/auto_deploy/singlegpu/smoke/test_ad_speculative_decoding.py": 209.38515135135134, - "unittest/auto_deploy/singlegpu/transformations": 333.13563862068963, - "unittest/auto_deploy/singlegpu/transformations/library/test_fuse_relu2_quant_nvfp4.py": 37.48995640326975, - "unittest/auto_deploy/singlegpu/transformations/library/test_moe_fusion.py": 66.10721253405994, - "unittest/auto_deploy/singlegpu/transformations/library/test_nvfp4_swiglu.py": 34.6551825613079, - "unittest/auto_deploy/singlegpu/transformations/library/test_quant_fusion.py": 40.50969945355191, - "unittest/auto_deploy/singlegpu/transformations/library/test_quant_moe.py": 32.301112021857925, - "unittest/auto_deploy/singlegpu/transformations/library/test_quantization.py": 33.53743989071038, - "unittest/auto_deploy/singlegpu/transformations/library/test_trtllm_gen_diag.py": 41.74485792349727, - "unittest/auto_deploy/singlegpu/utils": 45.690824827586205, - "unittest/auto_deploy/standalone": 749.0343914373088, - "unittest/bindings": 33.233741379310345, - "unittest/disaggregated/region/test_aux.py": 20.118003086419755, - "unittest/disaggregated/region/test_block.py": 23.155147471910112, - "unittest/disaggregated/region/test_page.py": 20.19467365269461, - "unittest/disaggregated/region/test_region.py": 20.12785579937304, - "unittest/disaggregated/test_agent_multi_backends.py": 82.44263218390805, - "unittest/disaggregated/test_bounce.py": 23.681536211699164, - "unittest/disaggregated/test_cache_reuse_adapter.py": 25.694189973614776, - "unittest/disaggregated/test_cache_transceiver_harness.py": 54.7884509202454, - "unittest/disaggregated/test_cache_transceiver_harness_report.py": 22.569489296636085, - "unittest/disaggregated/test_cache_transceiver_precheck_e2e.py": 102.7121677852349, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and no_window\"": 263.90550151975685, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and uniform_window\"": 232.48220388349515, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and no_window\"": 320.16261075949365, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and uniform_window\"": 325.7727388535032, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and vswa\"": 298.592413897281, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v1\"": 63.74947222222222, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v2\"": 79.69995151515151, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"no_window\"": 203.92816666666667, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"uniform_window\"": 208.466928125, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"v2 and vswa\"": 127.31946202531645, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_v1_dsa_indexer": 54.168414110429445, - "unittest/disaggregated/test_cluster_storage.py": 47.79166496163683, - "unittest/disaggregated/test_coordinator_e2e.py": 72.31078260869565, - "unittest/disaggregated/test_coordinator_worker.py": 32.91763541666666, - "unittest/disaggregated/test_deepseek_v4_kv_transfer.py": 456.1797496206373, - "unittest/disaggregated/test_disagg_cluster_manager_worker.py": 56.201279898218836, - "unittest/disaggregated/test_disagg_openai_client.py": 33.44577353689567, - "unittest/disaggregated/test_disagg_utils.py": 23.457208333333334, - "unittest/disaggregated/test_disaggregated_params.py": 20.980890675241156, - "unittest/disaggregated/test_extractor.py": 27.114717696629214, - "unittest/disaggregated/test_kv_transfer.py": 289.1068039215686, - "unittest/disaggregated/test_kv_transfer_mp.py": 84.64788253012048, - "unittest/disaggregated/test_mamba_transfer.py": 34.03104336734694, - "unittest/disaggregated/test_messenger.py": 23.249504201680672, - "unittest/disaggregated/test_minimax_m3_kv_transfer.py": 362.53847222222225, - "unittest/disaggregated/test_openai_disagg_server.py": 27.016676020408163, - "unittest/disaggregated/test_openai_disagg_service.py": 33.99901526717557, - "unittest/disaggregated/test_openai_server_info.py": 25.25883333333333, - "unittest/disaggregated/test_peer.py": 23.113104285714286, - "unittest/disaggregated/test_perf_logger.py": 20.46639936102236, - "unittest/disaggregated/test_pool_matching.py": 20.040252427184466, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first and not ctx_first_sync\"": 291.7294591836735, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first_sync\"": 289.61003000000005, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first1\"": 293.302887755102, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first2\"": 316.01223, - "unittest/disaggregated/test_rank_info.py": 20.087990936555894, - "unittest/disaggregated/test_remoteDictionary.py": 28.16403486750349, - "unittest/disaggregated/test_request_id.py": 20.111296072507553, - "unittest/disaggregated/test_router.py": 27.52228693181818, - "unittest/disaggregated/test_sync_transfer_timing.py": 20.093430420711975, - "unittest/disaggregated/test_transceiver_bounded_polling.py": 20.062371335504885, - "unittest/dynamo": 31.191930591259638, - "unittest/executor/test_base_worker.py": 132.89819083969468, - "unittest/executor/test_event_loop_error_broadcast.py": 23.20425503355705, - "unittest/executor/test_fatal_error_health_check.py": 27.783138539042824, - "unittest/executor/test_ipc.py": 28.95289724310777, - "unittest/executor/test_multi_frontend_routing.py": 20.487352668213457, - "unittest/executor/test_postprocessor_hook.py": 29.449583979328168, - "unittest/executor/test_proxy_fast_death.py": 30.67723711340206, - "unittest/executor/test_proxy_postproc_terminate.py": 25.31334010152284, - "unittest/executor/test_rpc.py": 434.0945229357798, - "unittest/executor/test_stats_serializer.py": 21.669697547683924, - "unittest/inputs/test_chat_template_dispatch.py": 26.911028645833333, - "unittest/inputs/test_content_format.py": 25.583408268733848, - "unittest/inputs/test_multimodal.py": 25.46571794871795, - "unittest/inputs/test_multimodal_input_processor.py": 25.350826873385014, - "unittest/inputs/test_url_validation.py": 25.648604651162792, - "unittest/inputs/test_video_decode.py": 25.576173126614986, - "unittest/kv_cache_manager_v2_tests": 228.90979301075268, - "unittest/llmapi/apps/_test_openai_chat_multimodal.py::test_single_chat_session_image_embeds -m needs_l40s": 157.64357537688443, - "unittest/llmapi/apps/_test_openai_embeddings.py": 154.04824111675129, - "unittest/llmapi/apps/_test_openai_multi_gpu.py -m \"part0\"": 121.33642857142857, - "unittest/llmapi/apps/test_chat_utils.py": 27.294929824561404, - "unittest/llmapi/apps/test_disagg_serving_perf_metrics.py": 80.77289215686274, - "unittest/llmapi/apps/test_encode_batcher.py": 27.280994923857868, - "unittest/llmapi/apps/test_harmony_channel_validation.py": 30.592439086294416, - "unittest/llmapi/apps/test_tool_parsers.py": 29.554723618090453, - "unittest/llmapi/test_additional_model_outputs.py -m \"gpu1\"": 62.16573945409429, - "unittest/llmapi/test_async_llm.py -m \"gpu2\"": 82.15391000000001, - "unittest/llmapi/test_async_llm.py -m \"gpu4\"": 88.93448484848484, - "unittest/llmapi/test_async_llm.py -m \"not (gpu2 or gpu4)\"": 196.92821637426903, - "unittest/llmapi/test_bench_async.py": 30.84113043478261, - "unittest/llmapi/test_deepseek_v4_tokenizer.py": 20.848953623188404, - "unittest/llmapi/test_embedding_arch_routing.py": 26.84065891472868, - "unittest/llmapi/test_executor.py": 29.101611111111108, - "unittest/llmapi/test_gc_utils.py": 31.859434343434344, - "unittest/llmapi/test_kv_cache_dtype_override.py": 26.816649122807018, - "unittest/llmapi/test_llm.py -m \"not part0\"": 225.55044444444445, - "unittest/llmapi/test_llm.py -m \"part0\"": 327.5163333333333, - "unittest/llmapi/test_llm_args.py": 132.44800253807108, - "unittest/llmapi/test_llm_encode.py": 64.248390625, - "unittest/llmapi/test_llm_encode_multi_item.py": 129.08888511749348, - "unittest/llmapi/test_llm_kv_cache_events.py::test_llm_api_attention_dp_kv_events": 54.56314285714286, - "unittest/llmapi/test_llm_multi_gpu_pytorch.py -m \"gpu4\"": 78.35294554455446, - "unittest/llmapi/test_llm_pytorch.py -m \"part0\"": 923.5839957507083, - "unittest/llmapi/test_llm_pytorch.py -m \"part1\"": 257.2634564315353, - "unittest/llmapi/test_llm_pytorch.py -m \"part2\"": 480.78776657458565, - "unittest/llmapi/test_llm_pytorch.py -m \"part3\"": 275.2785779310345, - "unittest/llmapi/test_llm_pytorch.py::test_nemotron_nas_lora": 247.856625, - "unittest/llmapi/test_llm_quant.py": 21.912051169590644, - "unittest/llmapi/test_llm_telemetry.py": 170.3385606060606, - "unittest/llmapi/test_llm_telemetry.py::TestTelemetryArchitectureExtraction": 71.03602319587628, - "unittest/llmapi/test_llm_telemetry.py::TestTelemetryPyTorchBackend": 71.8847385786802, - "unittest/llmapi/test_llm_utils.py": 30.531091836734692, - "unittest/llmapi/test_memory_profiling.py::test_pyexecutor_and_kvcache_share_execution_stream": 44.25662077922078, - "unittest/llmapi/test_mpi_session.py": 160.02251399491092, - "unittest/llmapi/test_mpi_session.py::test_llmapi_launch_multiple_tasks": 87.06310416666668, - "unittest/llmapi/test_reasoning_parser.py": 30.21788888888889, - "unittest/llmapi/test_request_priority.py": 25.484827067669173, - "unittest/llmapi/test_rlhf_utils.py": 25.711772727272727, - "unittest/llmapi/test_serialization.py": 25.378102301790282, - "unittest/llmapi/test_utils.py": 28.242190839694658, - "unittest/metrics/test_collector.py": 24.564216438356166, - "unittest/models/test_quant_config_utils.py": 25.456101265822785, - "unittest/others/test_cache_transceiver_precheck_config.py": 30.594468208092486, - "unittest/others/test_cache_transceiver_precheck_run.py": 27.6636449704142, - "unittest/others/test_convert_utils.py": 25.22678756476684, - "unittest/others/test_http_utils_fail_fast.py": 29.93428426395939, - "unittest/others/test_kv_cache_transceiver.py::test_async_transfer_keeps_llm_request_alive": 12.767763144717076, - "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mha]": 27.682636863136864, - "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mla]": 16.382363386613385, - "unittest/others/test_kv_cache_transceiver.py::test_cpp_nixl_sync_transfer_stress": 29.801800862068966, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[NIXL-mha-ctx_fp16_gen_fp16]": 23.806841658341657, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mha-ctx_fp16_gen_fp16]": 16.071417549167926, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mla-ctx_fp16_gen_fp16]": 15.999426494345718, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[UCX-mha-ctx_fp16_gen_fp16]": 12.542696803196803, - "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_silent_when_unset": 10.897987981972959, - "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_warns_once_per_request": 10.896794441662495, - "unittest/others/test_lm_eval.py": 28.92114, - "unittest/others/test_lora_manager.py": 35.860937984496125, - "unittest/others/test_lora_module_count.py": 27.002921319796954, - "unittest/others/test_time_breakdown.py": 27.481543589743588, - "unittest/others/test_tracing.py": 144.11626424870465, - "unittest/quantization/test_mode.py": 25.516699745547072, - "unittest/scaffolding": 395.9213879781421, - "unittest/scripts": 31.20939588688946, - "unittest/tools": 386.9344300518135, - "unittest/tools/test_layer_wise_benchmarks.py::test_deepseek_r1_ctx_dep[1]": 20.868882183908045, - "unittest/tools/test_layer_wise_benchmarks.py::test_nemotron_gen_dep[1]": 103.46749022164276, - "unittest/tools/test_layer_wise_benchmarks.py::test_performance_alignment[1]": 27.55558333333333, - "unittest/tools/test_layer_wise_benchmarks.py::test_qwen3_next_gen_tep[1]": 84.47367901234567, - "unittest/usage/test_collectors.py": 23.337350631136044, - "unittest/usage/test_config.py": 22.926307905686546, - "unittest/usage/test_e2e_capture.py": 22.975632911392402, - "unittest/usage/test_llmapi_config_telemetry_docs.py": 27.81584223918575, - "unittest/usage/test_opt_out.py": 22.998687065368568, - "unittest/usage/test_reporter.py": 26.33069496402878, - "unittest/usage/test_schema.py": 23.32576157082749, - "unittest/usage/test_transport.py": 29.48707735583685, - "unittest/utils/test_logger.py": 25.696051813471502, - "unittest/utils/test_util.py": 28.391819796954316, - "unittest/visual_gen/test_iteration_stats.py": 25.375496368038743, - "unittest/visual_gen/test_media_encoding.py": 25.67736519607843, - "unittest/visual_gen/test_output.py": 25.828207637231504, - "verl/test_verl_cases.py::test_async_generate": 681.7801428571429, - "verl/test_verl_cases.py::test_async_memory_management": 128.481, - "verl/test_verl_cases.py::test_init_without_device_mesh": 163.0692857142857, - "verl/test_verl_cases.py::test_inter_node_trtllm_rollout": 184.09014285714287, - "verl/test_verl_cases.py::test_make_async_request_get_method": 47.95542857142857, - "verl/test_verl_cases.py::test_make_async_request_http_error": 51.29528571428572, - "verl/test_verl_cases.py::test_make_async_request_max_attempts_exceeded": 50.296, - "verl/test_verl_cases.py::test_make_async_request_post_method": 45.08557142857143, - "verl/test_verl_cases.py::test_multimodal_different_image_sizes": 265.54571428571427, - "verl/test_verl_cases.py::test_multimodal_generate_with_image": 254.34057142857142, - "verl/test_verl_cases.py::test_multimodal_text_only_fallback": 215.1765714285714, - "verl/test_verl_cases.py::test_placement_group_multi_node_multi_replica": 48.83014285714285, - "verl/test_verl_cases.py::test_placement_group_multi_node_ray_resource_pool": 42.849, - "verl/test_verl_cases.py::test_placement_group_with_ray_resource_pool": 42.37528571428572, - "verl/test_verl_cases.py::test_placement_group_with_sub_ray_resource_pool": 41.65785714285715, - "verl/test_verl_cases.py::test_unimodal_batch_generate": 236.276, - "verl/test_verl_cases.py::test_unimodal_generate": 229.85028571428572, - "verl/test_verl_cases.py::test_wake_sleep_cycle": 193.665, - "visual_gen/test_visual_gen_benchmark.py::test_offline_benchmark": 90.47928571428571, - "visual_gen/test_visual_gen_benchmark.py::test_online_benchmark[openai-videos]": 91.72266666666667 + "accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 566.384125, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False]": 605.8393098591549, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 792.857375, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False]": 262.5967857142857, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True]": 158.5127857142857, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False]": 238.92535714285714, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True]": 161.73442857142857, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[adp-mtp2]": 311.97010309278346, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[noadp-mtp0]": 474.63128571428575, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_spec_dec": 178.47664285714288, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0]": 409.83045454545453, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2]": 744.224909090909, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 205.94233333333335, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 188.42866379310345, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV4Flash::test_auto_dtype": 592.6489204545454, + "accuracy/test_disaggregated_serving.py::TestGLM52NVFP4::test_nvfp4_nixl_python[cache_mgr_v1]": 990.9347659574468, + "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 342.7804705882353, + "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 415.1615625, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False]": 245.6970612244898, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True]": 247.0623829787234, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 255.7429157894737, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 424.93321978021976, + "accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4": 1390.96909375, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-False]": 208.08547872340426, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-True]": 209.256806122449, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-False]": 216.41998947368418, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-True]": 216.31725263157895, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-False]": 166.9120744680851, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-True]": 164.41655319148936, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-False]": 179.65035714285713, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-True]": 175.9757340425532, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-False]": 204.8307142857143, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-True]": 205.07693617021278, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-False]": 216.36536082474225, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-True]": 217.35315463917524, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-False]": 166.50253061224493, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-True]": 164.4506326530612, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-False]": 180.33906382978722, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-True]": 177.16786734693878, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_beam_search": 819.1626363636364, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_eagle3[eagle3_one_model=True-overlap_scheduler=True]": 171.46033653846152, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_gen_only_spec_dec": 217.6168762886598, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 76.94890816326532, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 156.0001489361702, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python": 520.4379043478261, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[GSM8K]": 175.96124, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[MMLU]": 102.43104, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp1]": 143.1969090909091, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp1]": 90.42032352941176, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=False]": 391.84766666666667, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=3-block_reuse=True-use_py_transceiver=False]": 478.8976666666667, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_ctx_dp2_gen_tp4": 281.545, + "accuracy/test_disaggregated_serving.py::TestQwen3NextInstruct::test_auto_dtype[use_py_transceiver=True]": 432.71708, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False]": 347.91988297872336, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-True]": 340.39185714285713, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[True-True]": 329.44937234042555, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2]": 326.08301020408163, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill": 348.02151020408166, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first": 1046.73604, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first_kv_cache_v1": 124.06744444444443, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend": 188.3538899082569, + "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_nvfp4]": 332.06218233618233, + "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[qwen3vl_2b_instruct]": 185.360093220339, + "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[no_overlap]": 69.15029714285714, + "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[overlap]": 67.1641367781155, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-ep2]": 177.65254545454545, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-tp2]": 293.10047959183674, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b]": 148.66536363636365, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[20b]": 130.28418181818182, + "accuracy/test_llm_api_autodeploy.py::TestGemma4MoE::test_bf16": 1040.9451304347826, + "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma3n_e2b_it": 652.3853028571428, + "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma4_e2b_it": 345.52021037463976, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_attention_dp[4]": 281.7136451612903, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[torch-True-1]": 909.5608181818183, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[triton-False-1]": 320.45450417827294, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-1]": 152.23809254143646, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-4]": 138.5165392670157, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-True-1]": 328.91990909090913, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[flashinfer]": 207.7224, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[trtllm]": 184.89575555555555, + "accuracy/test_llm_api_autodeploy.py::TestMiniMaxM2::test_finegrained_fp8": 607.0751460674157, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[Qwen_QwQ-32B-False]": 262.3381111111111, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[deepseek-ai_DeepSeek-R1-0528-True]": 809.233, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[google_gemma-3-1b-it-False]": 61.70166666666666, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.1-8B-Instruct-False]": 50.529333333333334, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.3-70B-Instruct-False]": 145.9032222222222, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Codestral-22B-v0.1-False]": 90.96744444444444, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Ministral-8B-Instruct-2410-False]": 64.63533333333334, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_DeepSeek-R1-0528-NVFP4-v2-True]": 1319.9854545454546, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-FP8-True]": 110.99866666666667, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-NVFP4-True]": 125.70889795918367, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-flashinfer_ssm-False]": 495.21763636363636, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-False]": 349.2198181818182, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-True]": 464.39036363636365, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-1-attn_dp_off-trtllm]": 1240.6362727272726, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 352.4914444444444, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-1-attn_dp_off-trtllm]": 393.3906777316736, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 259.3917204301075, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 219.00550537634408, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 315.08336363636363, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-4-attn_dp_off-trtllm]": 305.414, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 603.2348181818182, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 466.7068367346939, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 405.1242727272727, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 725.3176630727763, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-4-attn_dp_on-trtllm]": 510.6742727272727, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[bf16]": 240.1639674796748, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[fp8]": 225.47681818181817, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer]": 485.037, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm]": 375.2629090909091, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[fp8_ws4_80gb-trtllm]": 506.8681808510638, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[nvfp4_ws4_80gb-trtllm]": 512.3643636363637, + "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-4]": 1732.888907216495, + "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-8]": 2127.4628181818184, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[False]": 400.33763636363636, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[True]": 503.32218181818183, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_fp8[True]": 515.5300969101123, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_nvfp4[True]": 417.8357272727273, + "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_bf16_small[4]": 499.28675714285714, + "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_nvfp4[8]": 1178.8152727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[latency]": 1994.9886428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput]": 696.7412417218543, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency]": 815.1006, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp]": 445.61188888888887, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp_tp4]": 752.4664851485148, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen]": 924.8983000000001, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen_adp_lmtp]": 459.0964, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput]": 455.72643853820597, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_bs8_mtp]": 636.5326666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp]": 469.2359897260274, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_pp4_mtp]": 943.4426666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp4]": 888.5035, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8]": 409.552, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_dsa_host_cache_offload[host_cache_offload]": 960.41075, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline]": 899.9911666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline_mtp1]": 1043.9808039215686, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_attn_multi_gpus": 693.5356666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline]": 435.158900990099, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline_mtp1]": 384.64672277227726, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[baseline]": 424.88255208333334, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3::test_skip_softmax_attention_multi_gpus[target_sparsity_0.9]": 1936.0571020408163, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 109.189, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 111.4233, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 131.53244444444445, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 131.10222222222222, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 103.06677777777779, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 107.91711111111111, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 111.58439999999999, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 117.0749, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 147.24383333333336, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 150.32771428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 106.54339999999999, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 109.9482, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 129.39533333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 176.43553470437018, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 250.69174673629243, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 105.5962, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 111.16025, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 120.65477777777778, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 130.57385714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 100.0108, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 107.22133333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 185.2611, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 185.54422222222223, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 104.255, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 109.0158888888889, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 243.6549696969697, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=False]": 423.52147860962566, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=True]": 681.503816, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp": 216.84545161290322, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 75.69972727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 166.08079999999998, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 95.51, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 54.731818181818184, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.107, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 51.38036363636364, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.14818181818182, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 43.47654545454546, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 131.0839090909091, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 37.365199999999994, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 260.33008163265305, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[pp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 308.887675, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 54.127272727272725, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 47.00145454545454, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 44.096, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 48.436090909090915, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 50.10754545454546, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 50.03754545454546, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 61.07272727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 63.22081818181818, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 249.28828947368422, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 64.609, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 173.59127272727272, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 45.734, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 44.64572727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 54.545, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.67136363636364, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 69.7822, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 44.198, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 69.9719090909091, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 235.08795999999998, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 39.305, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 67.627, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=0]": 94.18960000000001, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=2]": 99.85809090909092, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=0]": 57.95333333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=2]": 70.299, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=0]": 72.82077777777778, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=2]": 74.54666666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=False]": 196.24513333333334, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=True]": 240.260800807537, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 257.00025, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=True]": 2899.535428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 242.9692307692308, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 1214.0928999999999, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 279.8845, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=True]": 3115.894142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 339.368, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 819.28775, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=fp8-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 435.4769372496662, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]": 214.80394305555555, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 227.8988691983122, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 294.3789699453552, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 279.15114520547945, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_dummy_load_format": 169.91534423897582, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 127.89110000000001, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 123.1895, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 106.7188, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 122.71088888888889, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 109.73866666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 122.89022222222222, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 104.65889999999999, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 201.32696816208394, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 107.83333333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 103.43469999999999, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 99.9161111111111, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 126.5578888888889, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 102.3408, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 105.42333333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 205.4267492063492, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 107.1272, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 127.02718181818183, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.298272727272725, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 48.119181818181815, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 129.86932352941176, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 72.26036363636364, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 46.696, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 73.61151960784314, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 65.76149019607843, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 98.89809090909091, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.159454545454544, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 66.94454545454546, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 51.37372727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 49.74481818181818, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 49.787454545454544, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 74.08118181818182, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 66.64172727272728, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 76.389, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 59.23609090909091, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 58.95272727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.89263636363636, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-sampler_async_worker=False]": 80.8312, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.64163636363636, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 53.18277777777778, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 134.4777777777778, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 67.16441176470589, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 61.92367647058824, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 113.32422222222222, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=True]": 115.1615925925926, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 51.46609090909091, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 66.509, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.761, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.95254545454546, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 134.4498148148148, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=0]": 68.2290737704918, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=2]": 154.16124725274724, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=0]": 51.68824509803922, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=2]": 60.39678217821782, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 79.26566666666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 223.61991005291003, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]": 110.20872727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=2]": 64.23452941176471, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=fp8-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 98.88074127906977, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 238.25687696335078, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 193.4495, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 325.0439166666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 110.1784, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 130.3401, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 108.9846, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.9392, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 119.118375, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 138.20675, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 117.565, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 158.50493922651935, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 121.72740331491713, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 172.87279999999998, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 171.1799, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 314.79475447570337, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.463, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 108.3047850340136, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 208.770375, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 201.36658333333335, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 189.09936363636365, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 417.9765, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 196.3797142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 236.29854545454543, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 119.75618181818183, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-pp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 246.47781818181818, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 155.82318181818184, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 382.13807936507936, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 164.28670588235295, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 140.84088888888888, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 55.83880952380952, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 109.41854545454545, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 229.1358, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 99.4054, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 131.42146153846153, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 123.13945454545454, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 114.74166666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 82.9781914893617, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 147.91820975609755, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 161.596, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 146.64896595744682, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 140.86234782608693, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 150.08861904761906, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 144.58642424242424, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 132.33410526315788, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 133.45738095238096, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 73.9634, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 127.70872727272727, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 177.698671875, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 73.15577570093458, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 211.7968248847926, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 203.5012, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=True-moe_backend=TRTLLM]": 132.16976041666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_batch_waiting[batch_wait_timeout_iters=10-batch_wait_max_tokens_ratio=1.0-mtp_nextn=0-fp8kv=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-v2_kv_cache=True]": 208.58932602739728, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=MEGAMOE_DEEPGEMM]": 737.5082040816327, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=TRTLLM]": 366.2747959183673, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy": 1131.2820365853659, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4ProDSpark::test_gsm8k_dep8_megamoe_deepgemm": 896.7543000000001, + "accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4[tp_size=8-ep_size=8]": 694.6597543859649, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dflash": 587.7711835616439, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format": 87.91589784946235, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[cutlass-one_model-overlap_scheduler]": 1955.2867352941175, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[triton-one_model-overlap_scheduler]": 1010.3446470588235, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-no_overlap_scheduler]": 688.9055999999999, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-overlap_scheduler]": 439.4791601941747, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-no_overlap_scheduler]": 426.94371428571424, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-overlap_scheduler]": 444.78338647342997, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-no_overlap_scheduler]": 550.945875, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-overlap_scheduler]": 630.1294747474748, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-no_overlap_scheduler]": 706.0050833333333, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-overlap_scheduler]": 591.5004950980392, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-one_model]": 280.22, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-two_model]": 398.28516666666667, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v2_kv_cache-one_model]": 1526.7824285714287, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-cutlass-auto]": 273.2275148648649, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-auto]": 241.6941, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-fp8]": 207.31851506849316, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-cutlass-auto]": 175.9200138121547, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-auto]": 248.3882, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-fp8]": 275.5822980501393, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-triton-auto]": 1710.5607717391304, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-auto]": 587.4411111111111, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-fp8]": 455.14378974358976, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-triton-auto]": 549.2571739130434, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-trtllm-auto]": 453.9397747747748, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-triton-auto]": 463.44741304347826, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-trtllm-auto]": 552.508797752809, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-triton-auto]": 900.5855666666666, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-auto]": 586.8257777777777, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-fp8]": 654.6051242937854, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-triton-auto]": 646.6367173913043, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-auto]": 508.7110540540541, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-fp8]": 469.1895714285714, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-tp4-trtllm-auto]": 634.4464065934066, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache_no_reuse-tp4-cutlass-auto]": 370.1528152173913, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus_online_eplb[fp8]": 282.8789893617021, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]": 327.9946086956522, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype": 201.60288607594936, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_reuse": 115.71518618618619, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_without_reuse": 128.66493072289157, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse": 49.06097674418604, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_disable_overlap_scheduler": 51.77953230769231, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_no_partial_reuse": 57.0330583090379, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_partial_reuse": 60.72586086956522, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_partial_reuse": 49.97132530120482, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse": 43.764580281690144, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_disable_overlap_scheduler": 47.694558641975306, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_low_memory_available": 57.55918452380953, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=False]": 183.97511111111112, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=True]": 188.8469, + "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_auto_dtype": 933.2313, + "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_fp8_prequantized": 419.3982222222223, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8]": 467.342, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8]": 657.7212, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8]": 1009.5471894736842, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8_attn_dp]": 682.5311111111112, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_bf16": 565.337822622108, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_fp8": 313.896358649789, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_nvfp4": 268.91661690885076, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_bf16_dflash": 331.3741212121212, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_fp8_dflash": 313.25575872093026, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_nvfp4_dflash": 356.3046358543417, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4": 148.77345046235138, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_4]": 37.77284679665738, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_64]": 136.69611111111112, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=False]": 87.9016, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=True]": 140.07416666666666, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=False]": 82.0882, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=True]": 114.42385714285714, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=FLASHINFER-torch_compile=False]": 134.89218181818183, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=TRTLLM-torch_compile=False]": 124.3952, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=FLASHINFER-torch_compile=False]": 46.56345454545455, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=False]": 41.564454545454545, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=True]": 46.043454545454544, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=False]": 86.29798181818181, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=True]": 68.56609090909092, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=TRTLLM-torch_compile=True]": 121.82825, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=FLASHINFER]": 157.36079999999998, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=TRTLLM]": 48.425628985507245, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=FLASHINFER]": 170.8326, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=TRTLLM]": 75.20133526011561, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dflash": 147.7249836065574, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dummy_load_format": 17.48164864864865, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=False-eagle3_one_model=True-overlap_scheduler=True]": 998.3113925501433, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=True-eagle3_one_model=True-overlap_scheduler=True]": 1011.6907043478261, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 74.59066666666668, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 110.4135, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 54.13452586206896, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 81.929431085044, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 73.4508888888889, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 110.1177, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 49.26115269461078, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 70.60236734693878, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[pp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 90.87263636363637, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 36.40436363636364, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 37.01018181818181, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 105.15461111111111, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 74.54427272727273, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 126.09972906403941, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 90.54714427860696, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 39.76645454545454, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 145.214, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 58.84855392156863, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 80.60036, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[llguidance]": 47.924777777777784, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 48.18807374631269, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[llguidance]": 47.427454545454545, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[llguidance-eagle3_one_model=True]": 58.863800000000005, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 51.88051724137931, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[llguidance]": 47.401900000000005, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[xgrammar]": 47.24155555555556, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=False]": 57.35074229691877, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=True]": 97.46340384615384, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=True-attn_backend=TRTLLM-torch_compile=True]": 179.64886426592798, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=False]": 785.3822138728324, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=True]": 745.1532246153846, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B_Instruct_RocketKV::test_auto_dtype": 924.9611, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 592.7475376344086, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=True]": 656.0788791208792, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 687.6400107526881, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=True]": 763.7198031914893, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_auto_dtype[tp_size=8-ep_size=8]": 10800.0, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_mxfp8[use_msa=False]": 10800.0, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_mxfp8_piecewise_cuda_graph[use_msa=False]": 10800.0, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=False]": 10800.0, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=True]": 1042.919387755102, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_fp8": 281.42, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_nvfp4_marlin_multi_gpus[tp_size=8]": 354.6746428571428, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-cpp_mamba_cache]": 373.539, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-python_mamba_cache]": 320.75507692307696, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-cpp_mamba_cache]": 214.66808333333336, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-python_mamba_cache]": 220.42576923076922, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar": 305.8700747663552, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar_custom_op": 212.21884615384616, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_OFF]": 408.7718888888889, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_ON]": 368.3760617977528, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[TEP4]": 630.2835833333334, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_hopper_w4a16": 509.27528571428576, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_online_eplb[moe_backend=TRTLLM]": 381.3987, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 341.6574444444444, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp": 666.2585544554455, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp_custom_op": 386.734, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3]": 838.8766666666667, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8]": 464.279, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_block_reuse[ADP4_MTP]": 1030.6662569832401, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 867.7993636363636, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_static_eplb[moe_backend=CUTLASS]": 470.4553846153846, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_off-trtllm]": 552.3911145833333, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_on-cutedsl]": 657.688081632653, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus": 2768.8230714285714, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_2gpu_mtp_ar": 198.87144444444445, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[dep4]": 341.21818181818185, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-cutlass]": 205.73988888888888, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-trtllm]": 178.89322222222222, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[tp1_block_reuse-cutlass]": 472.1311, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]": 230.0092222222222, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp]": 297.94870000000003, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_cutlass]": 1410.4496153846155, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_trtllm_eagle3]": 326.27490476190474, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_dummy_load_format": 45.34000611620795, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=False]": 151.40441666666666, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=True]": 188.0913354037267, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=False]": 127.05255555555556, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=True]": 157.40188888888886, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]": 138.786, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=True]": 176.966, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=False]": 137.82438095238095, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=True]": 125.6016, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=False]": 304.94709523809524, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=True]": 222.94661904761907, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=False]": 197.08683333333335, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=True]": 177.8886295264624, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=False]": 350.6162083333333, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=True]": 290.07529411764705, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=False]": 188.4275, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]": 222.8463482142857, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a16_mxfp4[latency-TRTLLM]": 99.26474238227146, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-CUTLASS]": 90.6996, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-TRTLLM]": 104.65530578512397, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-CUTLASS]": 85.41755555555557, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-TRTLLM]": 77.00845762711864, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention[target_sparsity_0.9-fp8kv=True]": 845.3932931506849, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=False]": 704.4539090909091, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=True]": 539.9865714285714, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False]": 615.3447894736842, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=True]": 547.9886875, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-CUTLASS]": 213.385125, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-TRTLLM]": 208.752, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-CUTLASS]": 225.09233333333333, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM]": 264.60998076923073, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_mtp": 355.00717987804876, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False]": 418.67988888888885, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_cutedsl]": 558.5049900990099, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_trtllm]": 438.426, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_block_reuse]": 627.075405940594, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_trtllm]": 475.7063333333333, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 948.6269726775956, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 569.6396262626263, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_mtp3_gdn_replay_tep4": 529.1704337349398, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_bf16": 188.67048780487804, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash": 260.3592507374631, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8": 190.37416863905324, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8_piecewise_cuda_graph": 267.21039432176656, + "accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM]": 225.70839265536725, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dflash": 166.4091335403727, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dummy_load_format": 39.61743076923077, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=False]": 100.84025382262996, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=True]": 118.26863636363636, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=True-enable_draft_len_schedule=False]": 113.90258333333333, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=True-enable_max_concurrency=False-enable_draft_len_schedule=False]": 198.20316666666665, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales[latency]": 205.6853557623479, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales_early_first_token_response": 116.59400867052022, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[fp8-latency]": 37.92944324324324, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[mxfp8-latency]": 53.90125, + "accuracy/test_llm_api_pytorch.py::TestSeedOss_36B::test_auto_dtype": 1173.8134285714286, + "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_auto_dtype[tp_size=8-ep_size=8]": 5400.0, + "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_fp8_block_scales[tp_size=4-ep_size=4-mtp_nextn=0]": 5400.0, + "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_fp8_block_scales[tp_size=4-ep_size=4-mtp_nextn=3]": 5400.0, + "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_nvfp4[tp_size=4-ep_size=4-mtp_nextn=0]": 5400.0, + "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_nvfp4[tp_size=4-ep_size=4-mtp_nextn=3]": 5400.0, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_cuda_graph_matches_eager_logits[tinyllama-1.1b]": 14.05920506329114, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[gemma-3-1b]": 19.97921739130435, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[phi-4-mini]": 22.435924242424242, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen2-7b]": 24.917961636828643, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen3-0.6b]": 9.642792307692307, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[starcoder2-3b]": 33.13732070707071, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[tinyllama-1.1b]": 6.582192405063291, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_cuda_graph_matches_eager_logits[bert-yelp]": 6.682479797979798, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-cuda_graph]": 4.394005050505051, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-eager]": 4.2008080808080805, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_per_token_reward[qwen2.5-prm-7b]": 22.808949494949495, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-0.6b]": 12.521669789227166, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-8b]": 33.73845667447307, + "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized": 325.59247987616095, + "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma4_26B_A4B::test_nvfp4": 374.86357142857145, + "accuracy/test_llm_api_pytorch_multimodal.py::TestKimiK25::test_nvfp4[dep8]": 2679.2065, + "accuracy/test_llm_api_pytorch_multimodal.py::TestMistralSmall24B::test_auto_dtype[forced_chunked_prefill]": 181.28377777777777, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[bf16]": 400.345, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8]": 690.9473907692308, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8_mmmu_encoder_cuda_graph]": 346.34280115273776, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]": 416.1755833333333, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype[forced_chunked_prefill]": 286.4858888888889, + "accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL::test_auto_dtype[forced_chunked_prefill]": 177.8486, + "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_fp8_block_scales[mtp_nextn=3]": 7200.0, + "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_nvfp4[mtp_nextn=0]": 7200.0, + "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_nvfp4[mtp_nextn=3]": 7200.0, + "cpp/test_multi_gpu.py::test_cache_transceiver[2proc-ucx_kvcache-90]": 178.4007340425532, + "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-nixl_kvcache-90]": 242.4074680851064, + "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-ucx_kvcache-90]": 275.1460652173913, + "cpp/test_multi_gpu.py::test_fused_gemm_allreduce[4proc-90]": 21.4115, + "cpp/test_multi_gpu.py::test_mpi_utils[90]": 220.0863829787234, + "cpp/test_multi_gpu.py::test_user_buffer[2proc-90]": 4.284468085106383, + "cpp/test_unit_tests.py::test_unit_tests[batch_manager-80]": 552.2707548209366, + "cpp/test_unit_tests.py::test_unit_tests[common-80]": 223.79825, + "cpp/test_unit_tests.py::test_unit_tests[common-90]": 222.75193593314765, + "cpp/test_unit_tests.py::test_unit_tests[executor-80]": 340.2329166666667, + "cpp/test_unit_tests.py::test_unit_tests[kernels-80]": 789.6245833333334, + "cpp/test_unit_tests.py::test_unit_tests[kernels-90]": 768.3216211699165, + "cpp/test_unit_tests.py::test_unit_tests[layers-80]": 913.9700833333334, + "cpp/test_unit_tests.py::test_unit_tests[layers-90]": 805.0914456824512, + "cpp/test_unit_tests.py::test_unit_tests[runtime-80]": 689.6249166666666, + "cpp/test_unit_tests.py::test_unit_tests[thop-80]": 3.291, + "cpp/test_unit_tests.py::test_unit_tests[thop-90]": 3.207640668523677, + "disaggregated/test_ad_disagg.py::test_async_eagle3_full_model_handoff": 97.89396703296703, + "disaggregated/test_ad_disagg.py::test_async_generation_matches_aggregate": 88.92743956043955, + "disaggregated/test_ad_disagg.py::test_async_generation_no_overlap_matches_aggregate": 87.01123076923076, + "disaggregated/test_ad_disagg.py::test_async_sharded_generation_handoff": 99.30441758241757, + "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[deepseek_v3_mla]": 161.1555747126437, + "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[tinyllama]": 98.48647550432277, + "disaggregated/test_ad_disagg.py::test_disaggregated_logits[deepseek_v3_mla]": 58.13549287749288, + "disaggregated/test_ad_disagg.py::test_disaggregated_logits[tinyllama]": 36.24735243553008, + "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[deepseek_v3_mla]": 142.15129971181557, + "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[tinyllama]": 97.21402017291066, + "disaggregated/test_ad_disagg.py::test_tinyllama_batch_handoff_semantic_slots": 77.39993123209169, + "disaggregated/test_ad_disagg_trtllm_serve.py::test_openai_completion": 62.068666666666665, + "disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin]": 107.51943, + "disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin]": 136.50441836734694, + "disaggregated/test_auto_scaling.py::test_minimal_instances[etcd-round_robin]": 105.13314285714286, + "disaggregated/test_auto_scaling.py::test_minimal_instances[http-round_robin]": 103.02789690721649, + "disaggregated/test_auto_scaling.py::test_service_discovery[etcd-round_robin]": 54.86334020618557, + "disaggregated/test_auto_scaling.py::test_service_discovery[http-kv_cache_aware]": 52.12690909090909, + "disaggregated/test_auto_scaling.py::test_service_discovery[http-round_robin]": 52.24916831683168, + "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-load_balancing]": 210.1801530612245, + "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-round_robin]": 207.23043564356436, + "disaggregated/test_auto_scaling.py::test_worker_restart[http-load_balancing]": 203.67193, + "disaggregated/test_disaggregated.py::test_disaggregated_benchmark_gen_only_insufficient_kv[TinyLlama-1.1B-Chat-v1.0]": 74.8770512195122, + "disaggregated/test_disaggregated.py::test_disaggregated_cache_aware_balance[TinyLlama-1.1B-Chat-v1.0]": 126.79705569007264, + "disaggregated/test_disaggregated.py::test_disaggregated_chat_completion_tool_calls[TinyLlama-1.1B-Chat-v1.0]": 70.9736107055961, + "disaggregated/test_disaggregated.py::test_disaggregated_conditional[TinyLlama-1.1B-Chat-v1.0]": 91.700578313253, + "disaggregated/test_disaggregated.py::test_disaggregated_cuda_graph[TinyLlama-1.1B-Chat-v1.0]": 81.07452153110047, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_conditional_v2[DeepSeek-V3-Lite-bf16]": 144.77602127659574, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_empty_batch[DeepSeek-V3-Lite-bf16]": 195.28455670103094, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 136.93694845360827, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 124.63738144329898, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 105.44671134020618, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 143.4193673469388, + "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 73.85774285714287, + "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 188.37085000000002, + "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 91.19978208232446, + "disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0]": 83.23244171779142, + "disaggregated/test_disaggregated.py::test_disaggregated_mixed[TinyLlama-1.1B-Chat-v1.0]": 131.9174964028777, + "disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_test[req120-conc64-qwen3_32b_fp8_mixed_stress]": 220.23792307692307, + "disaggregated/test_disaggregated.py::test_disaggregated_multi_gpu[TinyLlama-1.1B-Chat-v1.0]": 73.60878787878788, + "disaggregated/test_disaggregated.py::test_disaggregated_ngram[TinyLlama-1.1B-Chat-v1.0]": 87.51331400966183, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap[TinyLlama-1.1B-Chat-v1.0]": 102.21385019455253, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python[TinyLlama-1.1B-Chat-v1.0]": 83.16836734693877, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_bounce[TinyLlama-1.1B-Chat-v1.0]": 105.64901086956522, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_fabric_memory[TinyLlama-1.1B-Chat-v1.0]": 104.76987434554974, + "disaggregated/test_disaggregated.py::test_disaggregated_perf_metrics[TinyLlama-1.1B-Chat-v1.0]": 91.85413349514563, + "disaggregated/test_disaggregated.py::test_disaggregated_python_transceiver_host_offload[TinyLlama-1.1B-Chat-v1.0]": 72.75782105263158, + "disaggregated/test_disaggregated.py::test_disaggregated_qwen3_32b_fp8[Qwen3/Qwen3-32B-FP8]": 151.65707692307694, + "disaggregated/test_disaggregated.py::test_disaggregated_single_gpu[TinyLlama-1.1B-Chat-v1.0]": 86.56471394230769, + "disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0]": 71.0492895522388, + "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer[False-TinyLlama-1.1B-Chat-v1.0]": 54.64942857142857, + "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer_missing_blocks[False-TinyLlama-1.1B-Chat-v1.0]": 53.03601818181818, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_cancel_gen_requests[TinyLlama-1.1B-Chat-v1.0]": 33.09288200589971, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_llama_context_capacity[False-False-DeepSeek-V3-Lite-fp8/fp8]": 157.21709316770185, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[False-TinyLlama-1.1B-Chat-v1.0]": 46.077697058823524, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[True-TinyLlama-1.1B-Chat-v1.0]": 38.63458439716312, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[False-TinyLlama-1.1B-Chat-v1.0]": 43.92117916666667, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[True-TinyLlama-1.1B-Chat-v1.0]": 40.47734232954546, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8]": 101.05138360175695, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-True-DeepSeek-V3-Lite-fp8/fp8]": 89.19943502824859, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-False-DeepSeek-V3-Lite-fp8/fp8]": 104.34513333333334, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-True-DeepSeek-V3-Lite-fp8/fp8]": 104.84779656160458, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-False-TinyLlama-1.1B-Chat-v1.0]": 40.79223037323037, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-True-TinyLlama-1.1B-Chat-v1.0]": 44.52552798053528, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-False-TinyLlama-1.1B-Chat-v1.0]": 50.73347846889952, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-True-TinyLlama-1.1B-Chat-v1.0]": 50.999568345323745, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-False-Qwen3-8B-FP8]": 66.79170639534884, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-True-Qwen3-8B-FP8]": 61.497651162790696, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-False-Qwen3-8B-FP8]": 81.84019393939394, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-True-Qwen3-8B-FP8]": 63.21660349854228, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[False-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 57.31423822714682, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[True-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 56.641159292035404, + "disaggregated/test_workers.py::test_workers_conditional_disaggregation_deepseek_v3_lite_bf16[DeepSeek-V3-Lite-bf16]": 140.14826804123712, + "disaggregated/test_workers.py::test_workers_kv_cache_aware_router[TinyLlama-1.1B-Chat-v1.0]": 133.58280288461538, + "disaggregated/test_workers.py::test_workers_kv_cache_events[TinyLlama-1.1B-Chat-v1.0]": 79.16608937198068, + "examples/test_ad_guided_decoding.py::test_autodeploy_guided_decoding_main_json": 108.65893142857144, + "examples/test_ad_speculative_decoding.py::test_autodeploy_eagle3_one_model_acceptance_rate[flashinfer-torch-simple]": 77.57904189944134, + "examples/test_ad_speculative_decoding.py::test_eagle_model_with_weights": 2.5963277777777773, + "examples/test_ad_speculative_decoding.py::test_nemotron_mtp_model_with_weights": 835.9901866295264, + "examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke": 989.6935238095238, + "examples/test_llm_api_with_mpi.py::test_llm_api_single_gpu_with_mpirun[TinyLlama-1.1B-Chat-v1.0]": 49.392453516969994, + "examples/test_ray.py::test_llm_inference_async_ray": 129.92340111420614, + "examples/test_ray.py::test_llm_inference_distributed_ray[tep2]": 160.54990721649486, + "examples/test_ray.py::test_llm_inference_distributed_ray[tp2]": 68.30646391752578, + "examples/visual_gen/test_visual_gen.py::test_cosmos3_example": 297.64562113402064, + "examples/visual_gen/test_visual_gen.py::test_flux1_example": 248.61007180851064, + "examples/visual_gen/test_visual_gen.py::test_flux1_lpips_against_golden": 53.898, + "examples/visual_gen/test_visual_gen.py::test_flux2_example": 295.3289088541667, + "examples/visual_gen/test_visual_gen.py::test_flux2_lpips_against_golden": 103.39257142857143, + "examples/visual_gen/test_visual_gen.py::test_ltx2_cuda_graph_lpips_matches_eager": 338.59385714285713, + "examples/visual_gen/test_visual_gen.py::test_ltx2_example": 379.62811936339523, + "examples/visual_gen/test_visual_gen.py::test_ltx2_lpips_against_golden": 65.055, + "examples/visual_gen/test_visual_gen.py::test_qwen_image_example": 231.70423376623376, + "examples/visual_gen/test_visual_gen.py::test_qwenimage_cuda_graph_lpips_against_golden": 68.01414285714286, + "examples/visual_gen/test_visual_gen.py::test_qwenimage_lpips_against_golden": 65.81428571428572, + "examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough": 271.3475490196078, + "examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart": 199.75915289982424, + "examples/visual_gen/test_visual_gen.py::test_wan21_t2v_lpips_against_golden": 21.5448, + "examples/visual_gen/test_visual_gen.py::test_wan22_t2v_lpips_against_golden": 90.2427, + "examples/visual_gen/test_visual_gen.py::test_wan_t2v_example": 726.5599819121447, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[attn2d_2x2]": 839.0274545454546, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[cfg2_ulysses2]": 261.68454545454546, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[ulysses4]": 607.2706666666667, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[cfg2_tp2]": 236.3876, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2]": 71.3225, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_ulysses2]": 302.0615, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_batch_generation": 20.855231805929918, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_cold_start": 20.51119398907104, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_field_completeness": 12.331842541436465, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_full_block_reuse": 10.812967032967034, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_long_context": 38.267905913978495, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_partial_block_reuse": 19.251983516483516, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_rapid_fire": 14.49196694214876, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_shared_prefix": 20.32607988980716, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens": 333.74045871559633, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens": 132.6506574074074, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_eviction": 37.61371028037383, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_batch_size_limited": 107.80071629213484, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_block_reuse": 121.58346723646723, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_eviction_block_reuse": 20.056470752089137, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_multi_request": 129.70799719887955, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_with_eviction": 23.063224657534246, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[cuda_graph]": 24.279370473537604, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[no_cuda_graph]": 20.66685714285714, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_overlap": 17.89529213483146, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[non_overlap]": 123.73225210084034, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[overlap]": 121.00675637393768, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_partial_block_reuse": 121.48944350282486, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_token_budget_limited": 124.12047701149424, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_v2_vs_v1_basic": 128.52438797814207, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_chunked_prefill": 194.13237709497207, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_eviction": 58.65875284090909, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_multi_adapter_v2": 131.8038, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_v2": 112.23774785100287, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[guaranteed-chunked]": 159.27757306590257, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[max-util-chunked]": 159.43676802507838, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[no-overlap-chunked]": 170.64738461538462, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-chunked]": 166.18679824561406, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-no-chunked]": 164.12044207317075, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[python-scheduler]": 165.64470674486805, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[swa-chunked]": 133.17204360465118, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix_smoke": 133.89418328445748, + "llmapi/test_llm_api_connector.py::test_connector_async_onboard[False]": 18.619161835748795, + "llmapi/test_llm_api_connector.py::test_connector_async_onboard[True]": 18.2084039408867, + "llmapi/test_llm_api_connector.py::test_connector_async_save[False]": 15.075786924939468, + "llmapi/test_llm_api_connector.py::test_connector_async_save[True]": 12.830610294117648, + "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[False]": 25.233555282555283, + "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[True]": 24.342429611650484, + "llmapi/test_llm_api_connector.py::test_connector_e2e_persistent_cache": 22.890816625916873, + "llmapi/test_llm_api_connector.py::test_connector_multi_request": 10.410875305623472, + "llmapi/test_llm_api_connector.py::test_connector_priorities": 12.587395061728396, + "llmapi/test_llm_api_connector.py::test_connector_priorities_default": 12.447050724637682, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[attention_dp]": 4.182096153846154, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[beam_search]": 10.050549636803876, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[host_offloading]": 11.549893975903615, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[False]": 14.680273170731708, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[True]": 14.689616504854367, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[False]": 11.245905569007263, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[True]": 11.27094362745098, + "llmapi/test_llm_api_connector.py::test_connector_simple[False]": 12.478055288461539, + "llmapi/test_llm_api_connector.py::test_connector_simple[True]": 12.50396805896806, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-bart-large-cnn]": 26.003712328767122, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-bart-large-cnn]": 25.386357954545456, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-bart-large-cnn]": 22.717285714285715, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 7.3287808641975305, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-bart-large-cnn]": 6.780027777777778, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-bart-large-cnn]": 14.974052459016393, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-bart-large-cnn]": 22.9215, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 22.552, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 6.945199084668192, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 8.355032, + "llmapi/test_llm_api_pytorch_bart.py::test_mbart_pytorch_generate_encoder_decoder_end_to_end": 22.4349012345679, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_mixtral_moe_routed_expert_fp8_multi_lora_varying_ranks[cudagraph]": 5400.0, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_mixtral_moe_routed_expert_fp8_multi_lora_varying_ranks[eager]": 5400.0, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[cudagraph]": 79.19040469208211, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[eager]": 91.1888, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-t5-small]": 6.15115313225058, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 11.990981481481482, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-xl]": 38.03971428571428, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-t5-small]": 21.127428571428574, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-base]": 22.332, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-small]": 6.14457911908646, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-t5-small]": 12.96089082278481, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-t5-small]": 5.369179566563467, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-tp2-t5-small]": 17.288162393162395, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-off-greedy-byt5-small]": 13.017768595041321, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 21.729333333333333, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-t5-small]": 20.64085714285714, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-t5-small]": 5.132682098765432, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 21.11866666666667, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-t5-small]": 20.56142857142857, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 21.302833333333332, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-t5-small]": 20.521285714285714, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 20.4565, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-t5-small]": 19.902714285714286, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 20.3345, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-t5-small]": 19.869142857142858, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_context_generation_batch": 5.753551162790698, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-flan-t5-small]": 5.0157716535433075, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-t5-small]": 4.783761061946902, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-t5-small]": 5.329487640449438, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-t5-small]": 7.091325396825397, + "llmapi/test_llm_examples.py::test_llmapi_chat_example": 70.07263636363636, + "llmapi/test_llm_examples.py::test_llmapi_example_distributed_tp2": 56.021, + "llmapi/test_llm_examples.py::test_llmapi_example_guided_decoding": 65.55775636811832, + "llmapi/test_llm_examples.py::test_llmapi_example_inference": 71.85408210180624, + "llmapi/test_llm_examples.py::test_llmapi_example_inference_async": 53.34649383730485, + "llmapi/test_llm_examples.py::test_llmapi_example_inference_async_streaming": 48.074224548440064, + "llmapi/test_llm_examples.py::test_llmapi_example_logits_processor": 47.359884646962236, + "llmapi/test_llm_examples.py::test_llmapi_example_multilora": 61.29332004930156, + "llmapi/test_llm_examples.py::test_llmapi_kv_cache_connector[Qwen2-0.5B]": 91.71036363636364, + "llmapi/test_llm_examples.py::test_llmapi_quickstart": 48.46666379310345, + "llmapi/test_llm_examples.py::test_llmapi_quickstart_atexit": 319.7374347826087, + "llmapi/test_llm_examples.py::test_llmapi_server_example": 157.16363636363636, + "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_mtp": 244.0717337962963, + "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_ngram": 95.34129156908665, + "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_context": 1.0290494505494505, + "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_generation": 0.967483606557377, + "perf/host_perf/test_module_sampler.py::test_sampler_update_greedy[greedy_bs8]": 1.2933342618384402, + "perf/host_perf/test_module_sampler.py::test_sampler_update_stop_words[stopwords_bs32]": 1.4042594594594595, + "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_gen_only_bs8]": 0.925497191011236, + "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_mixed_32gen_4ctx]": 1.055345945945946, + "perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:8192]": 255.99544495412843, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 2016.2651538461537, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 379.92807692307696, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 1608.9949, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 951.954875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 371.3276923076923, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2365.190263414634, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 434.1525384615384, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 2853.0363846153846, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 376.83946153846153, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 2775.1226923076924, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 203.80992307692307, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 576.1800634146341, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1397.571923076923, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2195.1416153846153, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 501.6934615384616, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2228.081307317073, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 300.3468461538461, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 416.32678571428573, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 4561.989421052632, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2140.55235, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1500.5563333333332, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 2060.203, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1038.6396666666667, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 2913.0081578947365, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 423.3564375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 1640.4845625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1264.9850526315788, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1965.8842105263157, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 470.33705263157896, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 1331.24528125, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_tep8_mtp3]": 565.2403448275862, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 664.1381538461538, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 678.5075, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_dep4_mtp1_8k1k]": 916.9552524752474, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_1k8k]": 957.3329230769231, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_8k1k]": 573.0860742574257, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_1k8k]": 605.0051538461539, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 416.47730917874395, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_ad_blackwell-r1_fp8_ad_ws8_1k1k]": 755.3983636363637, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_blackwell-r1_fp8_tp8_6k1k]": 1130.9022307692308, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_dep8_mtp1_8k1k]": 817.33225, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_tep8_mtp3_8k1k]": 492.7644166666667, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_dep4_mtp1_8k1k]": 984.8252307692308, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_tep4_mtp3_8k1k]": 405.94523076923076, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_deepseek_v32_fp4_2_nodes_grace_blackwell-dsv32_fp4_dep8_trtllm_lpc_mnnvl_8k1k]": 498.78161538461535, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_adp_2k1k]": 574.2704166666666, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_235b_a22b_fp8_hopper-qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1k]": 957.4866363636363, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_32b_fp8_hopper-qwen3_32b_fp8_tp2_6k1k]": 776.424090909091, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 630.5806153846154, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_tep8_mtp3_8k1k]": 581.6817692307693, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 590.7949166666666, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 606.2082307692308, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_dep4_1k8k]": 1336.2756923076922, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tep2_1k8k]": 811.5083846153847, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 540.0438461538462, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp2_1k8k]": 395.9511538461539, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_deepseek_v3_lite-v3lite_fp8_bs8_128_256]": 573.4216694677872, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b-llama8b_fp16_bs8_128_256]": 268.8767824858757, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b_spec_decode-llama8b_spec_bs1_128_128]": 150.8482967032967, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_2_nodes_grace_blackwell-k25_thinking_fp4_dep8_32k8k]": 1488.879923076923, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_32k8k]": 1444.9784166666668, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_8k1k]": 766.74, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_32k8k]": 657.536, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_8k1k]": 445.1063846153846, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_dep4_8k1k]": 841.3804830917875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_tep4_8k1k]": 457.48476923076925, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_1000_1000]": 463.101, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_512_32]": 252.52316666666667, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]": 684.3395384615385, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]": 580.3056153846154, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep4_mtp3_8k1k]": 386.42491666666666, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tp4_8k1k]": 461.85966666666667, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_ad_blackwell-super_ad_ws4_1k1k]": 564.4200416666666, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_mtp_ad_blackwell-super_mtp_ad_ws4_1k1k]": 532.4755454545455, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 4133.825, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2472.7274545454547, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 3231.6324285714286, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 3142.5304615384616, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 851.2984308510638, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2151.5645, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2125.6215384615384, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 8068.51475, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2211.1548095238095, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1339.5592666666666, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 2621.412888888889, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1377.9775555555555, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1276.3097857142857, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 3251.7198439716312, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2548.912, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2466.507727272727, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 842.7718000000001, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 889.5297857142857, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 517.3953529411765, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 544.902, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL]": 883.5191538461539, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 312.7136666666667, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 264.12928571428574, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 828.4586666666667, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 488.6693076923077, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 3147.871157894737, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 836.2682272727272, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 1035.4665454545454, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 607.4167, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 586.0575, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1462.754, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 457.9464835680751, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-flux2_blackwell-flux2_fp8_cfg1_ulysses4_teacache_on]": 360.2670909090909, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_i2v_cfg2_ulysses4_compile_on]": 536.7069, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_t2v_cfg2_ulysses4_compile_on]": 474.23836363636366, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_nvfp4_i2v_cfg2_ulysses4_compile_on]": 458.0541818181818, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan21_t2v_14b_blackwell-wan21_14b_nvfp4_trtllm_cfg2_ulysses4_teacache_on]": 513.6001818181818, + "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp1_4instances]": 106.18771875, + "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp2_2instances]": 109.6773125, + "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-GUARANTEED_NO_EVICT-pytorch-stress-test]": 737.0806363636364, + "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-MAX_UTILIZATION-pytorch-stress-test]": 664.5234545454546, + "test_e2e.py::test_get_ci_container_port": 1.3634927884615387, + "test_e2e.py::test_openai_chat_example[pytorch]": 575.1747253012048, + "test_e2e.py::test_openai_chat_guided_decoding[meta-llama/Llama-3.1-8B-Instruct]": 74.3235283018868, + "test_e2e.py::test_openai_chat_guided_decoding[openai/gpt-oss-120b]": 183.251, + "test_e2e.py::test_openai_chat_harmony": 174.82321751412428, + "test_e2e.py::test_openai_chat_harmony_perf_metrics": 160.05857558139536, + "test_e2e.py::test_openai_chat_multimodal_example": 130.39714077669902, + "test_e2e.py::test_openai_chat_with_logit_bias[torch_sampler]": 86.90059294117647, + "test_e2e.py::test_openai_chat_with_logit_bias[trtllm_sampler]": 87.14738461538461, + "test_e2e.py::test_openai_completions_example[pytorch]": 261.67624220623503, + "test_e2e.py::test_openai_completions_with_logit_bias[torch_sampler]": 84.34349065420561, + "test_e2e.py::test_openai_completions_with_logit_bias[trtllm_sampler]": 84.66151643192488, + "test_e2e.py::test_openai_health": 72.27581818181818, + "test_e2e.py::test_openai_kv_cache_contamination": 1411.96477, + "test_e2e.py::test_openai_lora": 185.54133078880406, + "test_e2e.py::test_openai_misc_example[pytorch]": 350.13566343825664, + "test_e2e.py::test_openai_mmencoder_example": 136.0997242206235, + "test_e2e.py::test_openai_perf_metrics": 86.54126086956522, + "test_e2e.py::test_openai_post_processor": 372.86061686746984, + "test_e2e.py::test_openai_prometheus": 123.17233812949641, + "test_e2e.py::test_openai_reasoning[pytorch]": 257.26778132678135, + "test_e2e.py::test_openai_responses": 311.0094705882353, + "test_e2e.py::test_openai_responses_entrypoint": 107.0407848410758, + "test_e2e.py::test_openai_tool_call": 89.66446634615384, + "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-120B-gpt_oss/gpt-oss-120b]": 1067.7551818181819, + "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-20B-gpt_oss/gpt-oss-20b]": 214.3322994923858, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B]": 219.58391370558374, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]": 127.58688772845953, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta-Llama-3.1-8B]": 153.47955013192612, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]": 466.13843444730077, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_fp8_hf-Qwen3/saved_models_Qwen3-30B-A3B_fp8_hf]": 256.2311825192802, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_nvfp4_hf-Qwen3/saved_models_Qwen3-30B-A3B_nvfp4_hf]": 178.14921391752577, + "test_e2e.py::test_ptp_quickstart_advanced_bs1": 96.76857000000001, + "test_e2e.py::test_ptp_quickstart_advanced_deepseek_v3_lite_4gpus_adp_balance[DeepSeek-V3-Lite-FP8-DeepSeek-V3-Lite/fp8]": 94.38419387755103, + "test_e2e.py::test_ptp_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 80.07500583090379, + "test_e2e.py::test_ptp_quickstart_advanced_mixed_precision": 90.66226975476839, + "test_e2e.py::test_ptp_quickstart_advanced_mtp[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 190.38033875338755, + "test_e2e.py::test_ptp_quickstart_advanced_mtp_eagle[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 280.69667039106145, + "test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct]": 70.9769423631124, + "test_e2e.py::test_ptp_quickstart_bert[TRTLLM-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 23.737279936558288, + "test_e2e.py::test_ptp_quickstart_bert[VANILLA-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 7.569804761904762, + "test_e2e.py::test_trtllm_bench_help_sanity[meta-llama/Llama-3.1-8B]": 48.68448973607038, + "test_e2e.py::test_trtllm_bench_invalid_token_pytorch[TinyLlama-1.1B-Chat-v1.0-TinyLlama-1.1B-Chat-v1.0]": 79.17544525547444, + "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-non-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.94045479452055, + "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.49601497005989, + "test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b]": 120.57213636363636, + "test_e2e.py::test_trtllm_bench_mgmn": 110.8202, + "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-False-False]": 114.68200189393939, + "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-instruct-hf-fp8-True-True]": 69.32486930091186, + "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-]": 122.67552884615384, + "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-enable_request_rate]": 37.874408823529414, + "test_e2e.py::test_trtllm_serve_example": 178.86436363636363, + "test_e2e.py::test_trtllm_serve_lora_example": 94.39688040712468, + "test_e2e.py::test_trtllm_serve_multimodal_example": 178.93932289156626, + "test_e2e.py::test_trtllm_serve_top_logprobs[pytorch]": 86.90236144578313, + "test_fmha.py::test_fmha": 2348.44875, + "thirdparty/test_cmake_third_party.py::test_cmake_listfiles": 1.3952367149758453, + "thirdparty/test_git_modules.py::test_gitmodules": 1.2932296650717703, + "unittest/_torch/attention": 1499.0411369346734, + "unittest/_torch/attention --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1461.3758554216868, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_kernel.py": 26.301077142857142, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_module.py": 116.55956629834255, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_tf32.py": 21.41217094017094, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_cache_manager.py": 52.60678431372549, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_indices_transform.py": 50.76427808988764, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_o_proj.py": 37.86423978201635, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_sparse_mla.py": 52.061539944903586, + "unittest/_torch/attention/sparse/dsa/test_dsa_fp4_indexer.py": 26.329861386138614, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_cute_dsl_fp8_indexer_q_gemm_rope_fp4_matches_unfused": 45.63755555555556, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_dsv32_prefill_shape": 19.948702970297028, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_matches_deepgemm": 19.606653465346536, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_noncontiguous_split": 2.04908071278826, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_shape_dispatch": 19.43559405940594, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_indexer_k_cache_gather_contiguous_fp4": 19.109356435643562, + "unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py": 701.1912252747252, + "unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py": 348.6813176470588, + "unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1584.264129213483, + "unittest/_torch/attention/sparse/test_sparse_mla_forward.py": 111.27268644067796, + "unittest/_torch/attention/test_attention_mla.py": 135.2430928319624, + "unittest/_torch/auto_deploy": 26.348274924471298, + "unittest/_torch/compilation": 26.10724778761062, + "unittest/_torch/custom_ops/test_deepseek_v4_q_norm.py": 22.96618384401114, + "unittest/_torch/debugger": 20.013879710144927, + "unittest/_torch/distributed": 20.335233009708737, + "unittest/_torch/executor": 337.5190529327611, + "unittest/_torch/executor/test_async_transfer_manager.py": 27.127968215158926, + "unittest/_torch/executor/test_disagg_index_mapper_early_release.py": 27.426251781472686, + "unittest/_torch/executor/test_error_classification.py": 25.870426829268293, + "unittest/_torch/executor/test_hang_detector_kill.py": 42.569074018877295, + "unittest/_torch/executor/test_kv_cache_budget_split.py": 27.206930952380954, + "unittest/_torch/executor/test_kv_cache_compression_manager.py": 27.282539024390243, + "unittest/_torch/executor/test_kv_cache_estimation.py": 27.37214423076923, + "unittest/_torch/executor/test_kv_cache_v2_capacity_only.py": 25.55210606060606, + "unittest/_torch/executor/test_kv_pool_rebalance.py": 27.236201456310678, + "unittest/_torch/executor/test_model_loader_mx.py": 23.38812343201204, + "unittest/_torch/executor/test_overlap_scheduler.py": 796.4592737430168, + "unittest/_torch/executor/test_scheduler_serializable_output.py": 25.658269417475726, + "unittest/_torch/flashinfer/test_trtllm_flashinfer_symbol_collision.py": 1003.646782234957, + "unittest/_torch/lora": 21.104300771208226, + "unittest/_torch/memory": 20.41897507788162, + "unittest/_torch/misc": 217.8719063444109, + "unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy": 35.712514851485146, + "unittest/_torch/modeling -k \"modeling_gemma3\"": 250.2099910714286, + "unittest/_torch/modeling -k \"modeling_gpt_oss\"": 99.71302683615819, + "unittest/_torch/modeling -k \"modeling_llama\"": 124.79015498652292, + "unittest/_torch/modeling -k \"modeling_mixtral\"": 70.61155039787799, + "unittest/_torch/modeling -k \"modeling_nemotron_nano_v2_vl\"": 426.0704334140436, + "unittest/_torch/modeling -k \"modeling_nemotron_nas\"": 38.65103496503497, + "unittest/_torch/modeling -k \"modeling_out_of_tree\"": 146.20377011494253, + "unittest/_torch/modeling -k \"modeling_phi3\"": 33.42191375291375, + "unittest/_torch/modeling -k \"modeling_qwen\"": 707.234, + "unittest/_torch/modeling -k \"modeling_qwen_moe\"": 292.4790489510489, + "unittest/_torch/modeling -k \"modeling_siglip\"": 130.33033101851854, + "unittest/_torch/modeling -k \"modeling_speculative\"": 27.073209790209788, + "unittest/_torch/modeling -k \"modeling_vila\"": 42.617395784543326, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_multimodal_26b_dummy": 188.08098222222222, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_26b_dummy": 160.7302222222222, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_31b_dummy": 161.78340322580647, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy": 148.25198002853068, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[256]": 173.58835267857143, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[512]": 192.90066968325795, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e4b_dummy": 195.04240348330913, + "unittest/_torch/modeling/test_gemma4_multimodal.py": 24.872615803814714, + "unittest/_torch/modeling/test_modeling_afmoe.py": 65.08802259887005, + "unittest/_torch/modeling/test_modeling_cohere2.py": 60.729254196642685, + "unittest/_torch/modeling/test_modeling_cosmos3.py::TestCosmos3::test_all": 25.520618604651162, + "unittest/_torch/modeling/test_modeling_deepseekv4.py": 57.04133485193622, + "unittest/_torch/modeling/test_modeling_exaone_moe.py": 255.23339285714286, + "unittest/_torch/modeling/test_modeling_gemma4.py": 77.90781034482758, + "unittest/_torch/modeling/test_modeling_gemma4_unified.py": 20.19156862745098, + "unittest/_torch/modeling/test_modeling_llava_next.py::TestLlavaNext::test_all": 29.02741046511628, + "unittest/_torch/modeling/test_modeling_llava_next.py::test_llava_next_expand_prompt_token_ids_for_mm": 14.212421658986175, + "unittest/_torch/modeling/test_modeling_mistral.py": 37.79800717703349, + "unittest/_torch/modeling/test_modeling_nemotron_h.py::test_nemotron_h_sanity": 156.58529545454545, + "unittest/_torch/modeling/test_modeling_parakeet.py": 44.25659903381643, + "unittest/_torch/modeling/test_modeling_pixtral.py": 35.958952267303104, + "unittest/_torch/modeling/test_modeling_pixtral.py::test_tensor_parallelism": 43.275, + "unittest/_torch/modeling/test_modeling_qwen2_5vl.py::TestQwen2_5_VL::test_all": 76.14861520467836, + "unittest/_torch/modeling/test_modeling_qwen3_5_vl.py::TestQwen3_5VL::test_all": 73.7980761904762, + "unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py::TestQwen3_5MoeVL::test_all": 80.15000454545455, + "unittest/_torch/modeling/test_modeling_qwen3vl.py::TestQwen3VL::test_all": 96.30822105263158, + "unittest/_torch/modeling/test_modeling_qwen3vl.py::test_qwen3vl_init_preserves_caller_quant_config": 16.023488966318236, + "unittest/_torch/modeling/test_modeling_qwen3vl_moe.py::TestQwen3VLMoe::test_all": 82.7298538011696, + "unittest/_torch/modeling/test_modeling_radio.py": 29.436779904306217, + "unittest/_torch/modeling/test_modeling_step3p7.py": 28.5279375, + "unittest/_torch/modeling/test_modeling_step3p7vl.py": 38.826083720930235, + "unittest/_torch/modeling/test_multimodal_encoder_graph.py": 20.719585294117646, + "unittest/_torch/modeling/test_multimodal_encoder_mixin.py": 25.568744578313254, + "unittest/_torch/modeling/test_nemotron_nano_preprocessing.py": 62.31733732057416, + "unittest/_torch/models/checkpoints": 62.52584578313253, + "unittest/_torch/modules/dwdp/test_dwdp_fixup_moe_backends.py": 26.235854066985645, + "unittest/_torch/modules/dwdp/test_dwdp_manager.py": 25.706846715328467, + "unittest/_torch/modules/dwdp/test_dwdp_mapping.py": 25.72155339805825, + "unittest/_torch/modules/dwdp/test_dwdp_peer_ranges.py": 25.59992048192771, + "unittest/_torch/modules/fused_moe/test_deepgemm_fused_expand_quant.py": 22.875341658341657, + "unittest/_torch/modules/fused_moe/test_deepgemm_fused_gather_finalize.py": 22.920138532110094, + "unittest/_torch/modules/fused_ops/test_fused_qk_norm_rope_gate.py": 23.7237, + "unittest/_torch/modules/fused_ops/test_gelu_tanh_mul_fp4_quant.py": 24.07451566951567, + "unittest/_torch/modules/fused_ops/test_rmsnorm_fp4_quant.py": 23.92312912087912, + "unittest/_torch/modules/fused_ops/test_rmsnorm_residual_add.py": 24.953952802359883, + "unittest/_torch/modules/mamba": 740.4738101265823, + "unittest/_torch/modules/moe/test_cute_dsl_b12x_moe_backend.py": 21.06453807106599, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 91.753, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL\"": 342.06576923076926, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTLASS\"": 1208.6856129032258, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 46.524699999999996, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM\"": 62.843666666666664, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DENSEGEMM\"": 644.2885555555555, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_CUTEDSL\"": 243.494, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_DEEPGEMM\"": 27.79185714285714, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"TRTLLM\"": 123.4027619047619, + "unittest/_torch/modules/moe/test_moe_backend.py::test_trtllm_bf16_unquantized_moe": 899.1003994252874, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm": 206.84880319148937, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_boundary": 196.21933333333334, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_postquant": 118.7101182795699, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_nccl_ep_cuda_graph_replay_uses_updated_routing": 41.391349999999996, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTEDSL\"": 106.04984158415841, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES and not W4A8 and not MXFP8\"": 82.33142857142856, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES\"": 71.5973, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8_BLOCK_SCALES\"": 60.02572727272727, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and MXFP8 and not W4A8\"": 70.1191836734694, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and NVFP4\"": 86.05340816326532, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A16_MXFP4\"": 81.69265979381443, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_AWQ\"": 68.79285, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_FP8\"": 81.08657575757576, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_MXFP8\"": 150.906099009901, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W8A16\"": 91.86447959183673, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 84.99442574257425, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MARLIN\"": 81.10676923076923, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_CUTEDSL\"": 59.32976288659793, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_DEEPGEMM\"": 51.67359595959596, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and FP8_BLOCK_SCALES\"": 63.96949504950495, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and NVFP4 and not W4A8\"": 59.18147524752475, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A16_MXFP4\"": 87.15429591836735, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_MXFP4_MXFP8\"": 51.86418811881188, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_NVFP4_FP8\"": 157.12869306930693, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 64.47758, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 47.0703, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 54.87052, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 80.27069, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.16604, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_CUTEDSL-quant=NVFP4-routing=DeepSeekV3]": 55.19516326530612, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_DEEPGEMM-quant=W4A8_MXFP4_MXFP8-routing=DeepSeekV3]": 56.76075510204081, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 26.64341, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 26.58097, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 27.330119999999997, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 48.381660000000004, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.22708, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.49764, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb": 48.43913725490196, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_CUTEDSL\"": 56.218061224489794, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_DEEPGEMM\"": 52.125959183673466, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"not None\"": 25.63853608247423, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 46.70682, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 47.37518, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 47.52538, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 91.13814722222222, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and None\"": 27.01264705882353, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and not None\"": 151.68869058295965, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 49.13182647058823, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DENSEGEMM\"": 463.5970454545455, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_CUTEDSL\"": 175.77482905982907, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_DEEPGEMM\"": 42.02627173913043, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 65.25673630136987, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 32.85159831460675, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 30.820144329896905, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 29.589710743801653, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 30.251238754325257, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.18373224043716, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.542106529209626, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 33.902055401662054, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 29.093491408934707, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 57.66461690140845, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.324429906542058, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.float16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.324984813084114, + "unittest/_torch/modules/test_awq_quantization.py": 20.191866873065017, + "unittest/_torch/modules/test_engram.py": 102.07443352601156, + "unittest/_torch/modules/test_fp4_num_tokens_slice.py": 21.468339943342777, + "unittest/_torch/modules/test_fused_activation_quant.py": 21.324547932330827, + "unittest/_torch/modules/test_fused_add_rms_norm_quant.py": 21.579973869346734, + "unittest/_torch/modules/test_fused_rmsnorm_fp4_quantize.py": 28.061774853801168, + "unittest/_torch/modules/test_gemma4_fused_qkv_prep.py": 25.013674285714284, + "unittest/_torch/modules/test_group_rmn_norm.py": 26.816904549509367, + "unittest/_torch/modules/test_mhc.py": 29.68183914728682, + "unittest/_torch/modules/test_mla_helix.py": 52.47190936106984, + "unittest/_torch/modules/test_moe_host_sharer.py": 22.169184415584414, + "unittest/_torch/modules/test_moe_load_balancer.py": 22.058864864864862, + "unittest/_torch/modules/test_moe_routing.py": 47.47694806421152, + "unittest/_torch/modules/test_rotary_embedding.py": 22.013752016129033, + "unittest/_torch/modules/test_triton_linear.py": 21.429786885245903, + "unittest/_torch/modules/tests_lora_modules": 334.8057459807074, + "unittest/_torch/multi_gpu -m \"not post_merge\"": 340.56549333333334, + "unittest/_torch/multi_gpu_modeling -k \"deepseek\"": 311.1584347826087, + "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp1-bf16-trtllm-deepseekv3_lite]": 87.91341395348837, + "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp4-bf16-trtllm-deepseekv3_lite]": 115.07457209302325, + "unittest/_torch/multimodal -k \"not nemotron_nano_v2_vl_fp8\"": 914.8516534653465, + "unittest/_torch/multimodal/test_mm_encoder_standalone.py -k \"qwen3_30b_a3b_fp8\"": 386.13995362318843, + "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu2\"": 229.27255670103094, + "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu4\"": 470.16044791666667, + "unittest/_torch/ray_orchestrator/multi_gpu/test_inflight_weight_update.py": 83.6284947368421, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part0\"": 606.4082, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part1\"": 434.8369, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part2\"": 604.1570294117647, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part3\"": 521.5557475728156, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part4\"": 1015.58, + "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp1_4instances]": 278.0227083333333, + "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp2_2instances]": 293.1756666666667, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part0\"": 286.79657731958764, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part1\"": 287.2486494845361, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part2\"": 155.7070206185567, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part3\"": 156.41694845360826, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part4\"": 214.8349175257732, + "unittest/_torch/ray_orchestrator/single_gpu/test_cache_transceiver_comm.py": 50.02742076502732, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_sleep.py": 116.42223463687151, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part0\"": 718.5405393258427, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part1\"": 676.5004831460674, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part2\"": 496.98034550561795, + "unittest/_torch/sampler -k \"not test_speculative_d2h_parity_real_predictor\"": 629.0897239263804, + "unittest/_torch/sampler/test_beam_search.py": 466.1707026378897, + "unittest/_torch/sampler/test_beam_search_speculative_d2h.py": 130.66764, + "unittest/_torch/sampler/test_logits_logprobs.py": 274.3431100478469, + "unittest/_torch/sampler/test_torch_sampler.py": 207.59631818181816, + "unittest/_torch/sampler/test_trtllm_sampler.py": 98.30468292682926, + "unittest/_torch/speculative/hw_agnostic": 841.041305882353, + "unittest/_torch/speculative/test_eagle3.py": 657.948134218289, + "unittest/_torch/speculative/test_rejection_buffers_guard.py": 20.309, + "unittest/_torch/test_model_config.py": 25.657892523364488, + "unittest/_torch/test_torch_multi_arange.py": 46.24331719128329, + "unittest/_torch/thop/parallel": 1885.2997194670613, + "unittest/_torch/thop/parallel/test_finegrained_mixed_dtype_gemm.py": 28.596906801007556, + "unittest/_torch/thop/parallel/test_w4a16_linear.py": 21.82673299748111, + "unittest/_torch/thop/parallel/test_w4a8_linear.py": 20.717201511335013, + "unittest/_torch/thop/parallel/test_weight_only_quant_gemm.py": 58.659168765743075, + "unittest/_torch/thop/parallel/test_weight_only_quant_linear.py": 26.54437279596977, + "unittest/_torch/thop/parallel_hw_agnostic": 470.2435887573964, + "unittest/_torch/thop/serial": 52.232433867735466, + "unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py": 305.10693913043474, + "unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py": 218.1122830188679, + "unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py": 231.8527663551402, + "unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py": 192.40505319148937, + "unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py": 146.66659047619046, + "unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py": 70.97888679245283, + "unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py": 121.93633644859813, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py": 44.38675531914893, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py": 84.08998095238096, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py": 64.73582075471698, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py": 250.16000934579438, + "unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py": 206.29218279569892, + "unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py": 296.3244380952381, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py": 66.38530188679246, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py": 392.43442990654205, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py": 134.25890425531915, + "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py": 182.28520175438595, + "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_multinode.py": 22.10380172413793, + "unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py": 75.1850186915888, + "unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py": 308.29408510638297, + "unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py": 180.6913904761905, + "unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py": 455.3819038461539, + "unittest/_torch/visual_gen/test_attention_cute_dsl.py": 136.91421487603304, + "unittest/_torch/visual_gen/test_attention_cute_dsl_vsa.py": 43.35358474576271, + "unittest/_torch/visual_gen/test_attention_integration.py": 97.10202133333334, + "unittest/_torch/visual_gen/test_attention_perf.py": 210.67039617486338, + "unittest/_torch/visual_gen/test_attention_trtllm_sage.py": 189.85255145118734, + "unittest/_torch/visual_gen/test_cache_dit.py": 147.32329178470255, + "unittest/_torch/visual_gen/test_cosmos3_pipeline.py": 161.16777876106195, + "unittest/_torch/visual_gen/test_cosmos3_transformer.py": 98.51558529411764, + "unittest/_torch/visual_gen/test_flux_attention.py": 26.92808579088472, + "unittest/_torch/visual_gen/test_flux_infer.py": 25.655417233560094, + "unittest/_torch/visual_gen/test_flux_pipeline.py": 531.3778176795581, + "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxCombinedOptimizations::test_all_optimizations_combined": 77.22944036697247, + "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxParallelism::test_ulysses_2gpu_correctness": 64.36324778761062, + "unittest/_torch/visual_gen/test_flux_transformer.py": 38.82068067226891, + "unittest/_torch/visual_gen/test_ltx2_attention.py": 24.79130769230769, + "unittest/_torch/visual_gen/test_ltx2_pipeline.py": 717.9058804347826, + "unittest/_torch/visual_gen/test_ltx2_transformer.py": 27.601359173126614, + "unittest/_torch/visual_gen/test_model_loader.py": 46.429017441860466, + "unittest/_torch/visual_gen/test_quant_ops.py": 21.580876832844574, + "unittest/_torch/visual_gen/test_qwen_image_infer.py": 25.610972413793103, + "unittest/_torch/visual_gen/test_qwen_image_pipeline.py": 25.810387096774193, + "unittest/_torch/visual_gen/test_teacache.py": 20.69974858757062, + "unittest/_torch/visual_gen/test_tensor_payload.py": 25.807934240362812, + "unittest/_torch/visual_gen/test_trtllm_serve_e2e.py": 493.1424769647696, + "unittest/_torch/visual_gen/test_trtllm_serve_endpoints.py": 29.099208108108108, + "unittest/_torch/visual_gen/test_visual_gen_args.py": 21.59079120879121, + "unittest/_torch/visual_gen/test_visual_gen_params.py": 26.508512376237626, + "unittest/_torch/visual_gen/test_visual_gen_utils.py": 21.791359890109888, + "unittest/_torch/visual_gen/test_wan21_i2v_pipeline.py": 223.45977393617022, + "unittest/_torch/visual_gen/test_wan21_i2v_teacache.py": 348.9664619883041, + "unittest/_torch/visual_gen/test_wan21_t2v_pipeline.py": 192.33468047337277, + "unittest/_torch/visual_gen/test_wan21_t2v_teacache.py": 250.91146764705883, + "unittest/_torch/visual_gen/test_wan21_t2v_teacache_user_coefficients.py": 55.60930623306233, + "unittest/_torch/visual_gen/test_wan22_i2v_pipeline.py": 305.1618966480447, + "unittest/_torch/visual_gen/test_wan22_i2v_teacache.py": 149.2714253521127, + "unittest/_torch/visual_gen/test_wan22_t2v_pipeline.py": 287.8022845303867, + "unittest/_torch/visual_gen/test_wan22_t2v_teacache.py": 138.68507407407407, + "unittest/_torch/visual_gen/test_wan22_ti2v_5b_pipeline.py": 132.64411436170212, + "unittest/_torch/visual_gen/test_wan_transformer.py": 86.32018313953489, + "unittest/_torch/visual_gen/test_wan_vae.py": 66.85342215568862, + "unittest/_torch/visual_gen/test_wan_vsa_pipeline.py": 132.44576832844575, + "unittest/_torch/visual_gen/test_warmup.py": 21.973272237196763, + "unittest/_torch/weight_sharing": 25.870444964871197, + "unittest/api_stability": 29.708095823095825, + "unittest/auto_deploy/multigpu/compile": 43.853566666666666, + "unittest/auto_deploy/multigpu/smoke": 357.5653522012579, + "unittest/auto_deploy/multigpu/transformations": 2098.098430107527, + "unittest/auto_deploy/multigpu/transformations/library/test_allreduce_residual_rmsnorm_fusion.py -k \"strategy_auto\"": 119.75213402061856, + "unittest/auto_deploy/multigpu/transformations/library/test_tp_sharding.py::test_moe_tp_shard_nvfp4": 69.07451546391754, + "unittest/auto_deploy/singlegpu/compile": 68.55923552123552, + "unittest/auto_deploy/singlegpu/custom_ops": 522.1245855855856, + "unittest/auto_deploy/singlegpu/custom_ops/attention/test_triton_attention.py::TestSDPADispatch": 38.268032258064515, + "unittest/auto_deploy/singlegpu/custom_ops/mamba/test_flashinfer_mamba_cached_op.py": 47.28830913978495, + "unittest/auto_deploy/singlegpu/custom_ops/moe/test_ad_moe_op.py": 29.06793530997305, + "unittest/auto_deploy/singlegpu/custom_ops/moe/test_trtllm_moe.py": 40.64752291105122, + "unittest/auto_deploy/singlegpu/custom_ops/quantization/test_quant.py": 50.19106738544475, + "unittest/auto_deploy/singlegpu/mlir": 29.818291828793775, + "unittest/auto_deploy/singlegpu/models": 228.87805836575873, + "unittest/auto_deploy/singlegpu/shim": 91.26811240310077, + "unittest/auto_deploy/singlegpu/smoke": 1713.564464993395, + "unittest/auto_deploy/singlegpu/smoke/test_ad_build_small_single.py -k \"Nemotron-3-Nano-30B-A3B-FP8 or Nemotron-Nano-3-30B-A3.5B-dev or Llama-4-Scout\"": 249.2459164420485, + "unittest/auto_deploy/singlegpu/smoke/test_ad_speculative_decoding.py": 210.74064480874318, + "unittest/auto_deploy/singlegpu/transformations": 320.5240232858991, + "unittest/auto_deploy/singlegpu/transformations/library/test_fuse_relu2_quant_nvfp4.py": 36.78159029649596, + "unittest/auto_deploy/singlegpu/transformations/library/test_moe_fusion.py": 66.3970269541779, + "unittest/auto_deploy/singlegpu/transformations/library/test_nvfp4_swiglu.py": 34.473213513513514, + "unittest/auto_deploy/singlegpu/transformations/library/test_quant_fusion.py": 39.26718699186991, + "unittest/auto_deploy/singlegpu/transformations/library/test_quant_moe.py": 31.283571815718158, + "unittest/auto_deploy/singlegpu/transformations/library/test_quantization.py": 33.57898641304348, + "unittest/auto_deploy/singlegpu/transformations/library/test_trtllm_gen_diag.py": 41.16601630434782, + "unittest/auto_deploy/singlegpu/utils": 45.979049222797926, + "unittest/auto_deploy/standalone": 719.8760111731843, + "unittest/bindings": 41.77800726392252, + "unittest/disaggregated/region/test_aux.py": 20.04190855457227, + "unittest/disaggregated/region/test_block.py": 23.17737385620915, + "unittest/disaggregated/region/test_page.py": 20.239856756756755, + "unittest/disaggregated/region/test_region.py": 20.012345047923322, + "unittest/disaggregated/test_agent_multi_backends.py": 82.98323180592992, + "unittest/disaggregated/test_bounce.py": 23.589685828877005, + "unittest/disaggregated/test_cache_reuse_adapter.py": 25.585138888888892, + "unittest/disaggregated/test_cache_transceiver_harness.py": 56.23020353982301, + "unittest/disaggregated/test_cache_transceiver_harness_report.py": 21.828332378223493, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and no_window\"": 189.49816756756758, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and uniform_window\"": 171.13203492063494, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and no_window\"": 228.43785139318885, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and uniform_window\"": 243.70949554896143, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and vswa\"": 221.75007692307693, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v1\"": 53.39233522727273, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v2\"": 64.50010810810811, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"no_window\"": 163.61731012658228, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"uniform_window\"": 150.50420743034056, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"v2 and vswa\"": 102.53459940652819, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_v1_dsa_indexer": 49.56948913043478, + "unittest/disaggregated/test_cluster_storage.py": 47.17109420289855, + "unittest/disaggregated/test_coordinator_e2e.py": 72.26801550387597, + "unittest/disaggregated/test_coordinator_worker.py": 32.890360103626946, + "unittest/disaggregated/test_deepseek_v4_kv_transfer.py": 530.9268578431372, + "unittest/disaggregated/test_disagg_cluster_manager_worker.py": 55.52555365853659, + "unittest/disaggregated/test_disagg_openai_client.py": 33.3553485576923, + "unittest/disaggregated/test_disagg_utils.py": 23.560001321003963, + "unittest/disaggregated/test_disaggregated_params.py": 21.249914201183433, + "unittest/disaggregated/test_extractor.py": 26.12241229193342, + "unittest/disaggregated/test_kv_transfer.py": 607.4314255319149, + "unittest/disaggregated/test_kv_transfer_mp.py": 104.17107902735562, + "unittest/disaggregated/test_mamba_transfer.py": 33.60644819277109, + "unittest/disaggregated/test_messenger.py": 23.172675066312998, + "unittest/disaggregated/test_minimax_m3_kv_transfer.py": 337.1307835051547, + "unittest/disaggregated/test_openai_disagg_server.py": 27.25944660194175, + "unittest/disaggregated/test_openai_disagg_service.py": 33.72867386091127, + "unittest/disaggregated/test_openai_server_info.py": 28.47416666666667, + "unittest/disaggregated/test_peer.py": 23.180986541049798, + "unittest/disaggregated/test_perf_logger.py": 20.14458017492711, + "unittest/disaggregated/test_pool_matching.py": 19.757593406593404, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first and not ctx_first_sync\"": 442.66612371134016, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first_sync\"": 409.2202551020408, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first1\"": 446.7769587628866, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first2\"": 422.67712244897956, + "unittest/disaggregated/test_rank_info.py": 19.967412742382272, + "unittest/disaggregated/test_remoteDictionary.py": 28.488994572591587, + "unittest/disaggregated/test_request_id.py": 19.980398791540786, + "unittest/disaggregated/test_router.py": 25.549286479250334, + "unittest/disaggregated/test_sync_transfer_timing.py": 19.801176470588235, + "unittest/disaggregated/test_transceiver_bounded_polling.py": 20.033817204301073, + "unittest/dynamo": 30.93964938271605, + "unittest/executor/test_base_worker.py": 131.1828982630273, + "unittest/executor/test_fatal_error_health_check.py": 28.003123831775703, + "unittest/executor/test_ipc.py": 29.161146853146853, + "unittest/executor/test_multi_frontend_routing.py": 20.379634020618557, + "unittest/executor/test_postprocessor_hook.py": 29.60544927536232, + "unittest/executor/test_proxy_fast_death.py": 30.611908536585364, + "unittest/executor/test_proxy_postproc_terminate.py": 25.622395683453238, + "unittest/executor/test_rpc.py": 428.50881176470585, + "unittest/executor/test_stats_serializer.py": 22.270259154929576, + "unittest/inputs/test_chat_template_dispatch.py": 27.11566503667482, + "unittest/inputs/test_content_format.py": 25.79082608695652, + "unittest/inputs/test_multimodal.py": 25.693597087378638, + "unittest/inputs/test_multimodal_input_processor.py": 25.525548309178742, + "unittest/inputs/test_url_validation.py": 25.912739659367396, + "unittest/inputs/test_video_decode.py": 25.766543046357615, + "unittest/kv_cache_manager_v2_tests": 213.1598582781457, + "unittest/llmapi/apps/_test_openai_chat_multimodal.py::test_single_chat_session_image_embeds -m needs_l40s": 160.2151168224299, + "unittest/llmapi/apps/_test_openai_embeddings.py": 156.40381323877068, + "unittest/llmapi/apps/_test_openai_multi_gpu.py -m \"part0\"": 119.85816666666668, + "unittest/llmapi/apps/test_chat_utils.py": 27.616254716981132, + "unittest/llmapi/apps/test_disagg_serving_perf_metrics.py": 78.61523157894736, + "unittest/llmapi/apps/test_encode_batcher.py": 27.48177108433735, + "unittest/llmapi/apps/test_harmony_channel_validation.py": 30.0566327014218, + "unittest/llmapi/apps/test_tool_parsers.py": 29.678218009478673, + "unittest/llmapi/test_additional_model_outputs.py -m \"gpu1\"": 61.69495260663508, + "unittest/llmapi/test_async_llm.py -m \"gpu2\"": 83.43303092783505, + "unittest/llmapi/test_async_llm.py -m \"gpu4\"": 89.11570833333333, + "unittest/llmapi/test_async_llm.py -m \"not (gpu2 or gpu4)\"": 201.0308156424581, + "unittest/llmapi/test_deepseek_v4_tokenizer.py": 20.679311764705883, + "unittest/llmapi/test_embedding_arch_routing.py": 26.95822033898305, + "unittest/llmapi/test_executor.py": 27.13577777777778, + "unittest/llmapi/test_gc_utils.py": 32.136959427207636, + "unittest/llmapi/test_kv_cache_dtype_override.py": 27.04092523364486, + "unittest/llmapi/test_llm.py -m \"not part0\"": 214.32133333333334, + "unittest/llmapi/test_llm.py -m \"part0\"": 323.924, + "unittest/llmapi/test_llm_args.py": 131.87056698564595, + "unittest/llmapi/test_llm_encode.py": 61.49556812339331, + "unittest/llmapi/test_llm_encode_multi_item.py": 126.70357742782153, + "unittest/llmapi/test_llm_kv_cache_events.py::test_llm_api_attention_dp_kv_events": 53.872166666666665, + "unittest/llmapi/test_llm_multi_gpu_pytorch.py -m \"gpu4\"": 72.83206930693069, + "unittest/llmapi/test_llm_pytorch.py -m \"part0\"": 919.85638593482, + "unittest/llmapi/test_llm_pytorch.py -m \"part1\"": 258.6908440860215, + "unittest/llmapi/test_llm_pytorch.py -m \"part2\"": 516.3453256445048, + "unittest/llmapi/test_llm_pytorch.py -m \"part3\"": 280.88529755434786, + "unittest/llmapi/test_llm_pytorch.py::test_nemotron_nas_lora": 197.97414285714288, + "unittest/llmapi/test_llm_quant.py": 27.24870817843866, + "unittest/llmapi/test_llm_telemetry.py": 175.43786629526463, + "unittest/llmapi/test_llm_telemetry.py::TestTelemetryArchitectureExtraction": 72.02631630170316, + "unittest/llmapi/test_llm_telemetry.py::TestTelemetryPyTorchBackend": 71.87491931540343, + "unittest/llmapi/test_llm_utils.py": 31.80273798076923, + "unittest/llmapi/test_memory_profiling.py::test_pyexecutor_and_kvcache_share_execution_stream": 41.98156234096692, + "unittest/llmapi/test_mpi_session.py": 156.37610396039605, + "unittest/llmapi/test_mpi_session.py::test_llmapi_launch_multiple_tasks": 88.6355319148936, + "unittest/llmapi/test_reasoning_parser.py": 29.937814285714285, + "unittest/llmapi/test_request_priority.py": 25.679375296912113, + "unittest/llmapi/test_serialization.py": 25.607498806682578, + "unittest/llmapi/test_utils.py": 27.70747016706444, + "unittest/metrics/test_collector.py": 23.620726519337016, + "unittest/models/test_quant_config_utils.py": 25.616273809523808, + "unittest/others/test_convert_utils.py": 25.440990243902437, + "unittest/others/test_kv_cache_transceiver.py::test_async_transfer_keeps_llm_request_alive": 12.929715589541194, + "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mha]": 28.022305486900642, + "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mla]": 16.43778546712803, + "unittest/others/test_kv_cache_transceiver.py::test_cpp_nixl_sync_transfer_stress": 31.1141288945796, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[NIXL-mha-ctx_fp16_gen_fp16]": 24.296445485939813, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mha-ctx_fp16_gen_fp16]": 25.86616713091922, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mla-ctx_fp16_gen_fp16]": 24.195852528089887, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[UCX-mha-ctx_fp16_gen_fp16]": 12.57014578194376, + "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_silent_when_unset": 11.213987419832264, + "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_warns_once_per_request": 11.21132955106068, + "unittest/others/test_lora_manager.py": 36.074141463414634, + "unittest/others/test_lora_module_count.py": 27.08678883495146, + "unittest/others/test_time_breakdown.py": 27.776176904176904, + "unittest/others/test_tracing.py": 147.28475609756097, + "unittest/quantization/test_mode.py": 25.56721153846154, + "unittest/scaffolding": 446.7132156862745, + "unittest/scripts": 31.32473975903614, + "unittest/tools": 376.44609876543205, + "unittest/tools/test_layer_wise_benchmarks.py::test_deepseek_r1_ctx_dep[1]": 20.500902515723272, + "unittest/tools/test_layer_wise_benchmarks.py::test_nemotron_gen_dep[1]": 114.82735611510792, + "unittest/tools/test_layer_wise_benchmarks.py::test_performance_alignment[1]": 29.439571428571426, + "unittest/tools/test_layer_wise_benchmarks.py::test_qwen3_next_gen_tep[1]": 88.21061647058823, + "unittest/usage/test_collectors.py": 23.418028416779432, + "unittest/usage/test_config.py": 23.08291386271871, + "unittest/usage/test_e2e_capture.py": 23.17349802890933, + "unittest/usage/test_llmapi_config_telemetry_docs.py": 28.078541871921182, + "unittest/usage/test_opt_out.py": 23.03156282722513, + "unittest/usage/test_reporter.py": 26.090890645586295, + "unittest/usage/test_schema.py": 23.357160167130917, + "unittest/usage/test_transport.py": 29.417058265582654, + "unittest/utils/test_logger.py": 25.89497136038186, + "unittest/utils/test_util.py": 28.75711650485437, + "unittest/visual_gen/test_iteration_stats.py": 25.598708624708625, + "unittest/visual_gen/test_media_encoding.py": 25.9517546728972, + "unittest/visual_gen/test_output.py": 25.937927272727272, + "verl/test_verl_cases.py::test_async_generate": 424.8718181818182, + "verl/test_verl_cases.py::test_async_memory_management": 93.11290909090908, + "verl/test_verl_cases.py::test_init_without_device_mesh": 119.52972727272729, + "verl/test_verl_cases.py::test_inter_node_trtllm_rollout": 121.82945454545454, + "verl/test_verl_cases.py::test_make_async_request_get_method": 24.27409090909091, + "verl/test_verl_cases.py::test_make_async_request_http_error": 24.187363636363635, + "verl/test_verl_cases.py::test_make_async_request_max_attempts_exceeded": 24.261909090909093, + "verl/test_verl_cases.py::test_make_async_request_post_method": 25.87009090909091, + "verl/test_verl_cases.py::test_multimodal_different_image_sizes": 171.35136363636366, + "verl/test_verl_cases.py::test_multimodal_generate_with_image": 127.19154545454546, + "verl/test_verl_cases.py::test_multimodal_text_only_fallback": 115.52581818181818, + "verl/test_verl_cases.py::test_placement_group_multi_node_multi_replica": 21.758363636363637, + "verl/test_verl_cases.py::test_placement_group_multi_node_ray_resource_pool": 21.58818181818182, + "verl/test_verl_cases.py::test_placement_group_with_ray_resource_pool": 21.164454545454543, + "verl/test_verl_cases.py::test_placement_group_with_sub_ray_resource_pool": 21.24172727272727, + "verl/test_verl_cases.py::test_unimodal_batch_generate": 128.48545454545453, + "verl/test_verl_cases.py::test_unimodal_generate": 103.0079090909091, + "verl/test_verl_cases.py::test_wake_sleep_cycle": 106.03390909090909, + "visual_gen/test_visual_gen_benchmark.py::test_offline_benchmark": 74.4884, + "visual_gen/test_visual_gen_benchmark.py::test_online_benchmark[openai-videos]": 96.1846 } diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index e79dfe4b7318..8f6a699f9c7c 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1121,93 +1121,6 @@ def test_gen_only_spec_dec(self): tensor_parallel_size=4) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - @skip_pre_blackwell - @pytest.mark.skip_less_device(8) - @pytest.mark.parametrize("gen_pp,gen_tp,gen_cp,enable_attention_dp", [ - (1, 2, 2, False), - (1, 2, 2, True), - ], - ids=["pp1tp2cp2", "pp1dp2cp2"]) - @pytest.mark.parametrize("cuda_graph_config", [ - { - "enable_padding": True, - "batch_sizes": [1, 2, 4, 8, 16, 32, 64] - }, - ], - ids=["cudagraph:with_padding"]) - @pytest.mark.parametrize("comms_medium", ["fifo_v2"]) - def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config, - gen_pp, gen_tp, gen_cp, enable_attention_dp): - # Parse comms_medium to get use_nccl_for_alltoall and fifo_version. - if comms_medium == "nccl": - use_nccl_for_alltoall = True - fifo_version = 2 # Not used when NCCL is enabled. - elif comms_medium == "fifo_v1": - use_nccl_for_alltoall = False - fifo_version = 1 - elif comms_medium == "fifo_v2": - use_nccl_for_alltoall = False - fifo_version = 2 - else: - raise ValueError(f"Unknown comms_medium: {comms_medium}") - gen_ep = gen_tp * gen_cp - kv_cache_config = { - "free_gpu_memory_fraction": 0.5, - "enable_block_reuse": False, - "enable_partial_reuse": False, - "tokens_per_block": 32, - } - ctx_server_config = { - "pipeline_parallel_size": 1, - "tensor_parallel_size": 4, - "context_parallel_size": 1, - "disable_overlap_scheduler": True, - "kv_cache_config": kv_cache_config, - "enable_chunked_prefill": False, - "cuda_graph_config": None, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 8192, - }, - } - gen_server_config = { - "tensor_parallel_size": gen_tp, - "pipeline_parallel_size": gen_pp, - "context_parallel_size": gen_cp, - "moe_expert_parallel_size": gen_ep, - "cp_config": { - "cp_type": "HELIX", - "tokens_per_block": 32, - "use_nccl_for_alltoall": use_nccl_for_alltoall, - "fifo_version": fifo_version, - }, - "disable_overlap_scheduler": True, - "kv_cache_config": kv_cache_config, - "enable_chunked_prefill": False, - "cuda_graph_config": cuda_graph_config, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 8192, - }, - "enable_attention_dp": enable_attention_dp, - } - disaggregated_server_config = { - "hostname": "localhost", - "backend": "pytorch", - "context_servers": { - "num_instances": 1 - }, - "generation_servers": { - "num_instances": 1 - } - } - with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config, gen_server_config, - self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - @pytest.mark.skip_less_device(2) @pytest.mark.skip_less_device_memory(60000) @parametrize_with_ids("mtp_nextn", [0, 2]) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index aec114562a42..9faf48b249cf 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,7 +1,5 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml index 762bda05cdf6..e2d536755fdc 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml @@ -160,7 +160,6 @@ l0_dgx_b200: backend: pytorch orchestrator: mpi tests: - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] TIMEOUT (60) - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp] TIMEOUT (60) @@ -182,8 +181,8 @@ l0_dgx_b200: - accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4_mtp_index_share[tp_size=8-ep_size=8] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy TIMEOUT (240) - examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke TIMEOUT (120) - - accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4 TIMEOUT (180) - accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] TIMEOUT (60) + - accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4 TIMEOUT (180) - accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekV4ProDSpark::test_gsm8k_dep8_megamoe_deepgemm TIMEOUT (240) - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp TIMEOUT (60) diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index c8883de9e597..69a6cef9cb18 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -170,8 +170,11 @@ full:B200/llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_m full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/Qwen3-235B-A22B-tp16-mmlu] SKIP (https://nvbugs/6424188) full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/saved_models_Qwen3-235B-A22B_nvfp4_hf-tp16-mmlu] SKIP (https://nvbugs/6424188) full:B200/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) +<<<<<<< HEAD full:B300/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] SKIP (https://nvbugs/6410881) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6539942) +======= +>>>>>>> 5a1acec118 (remove TestDeepSeekV3Lite::test_auto_dtype_with_helix as helix can't run with Python transceiver) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6483370) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer] SKIP (https://nvbugs/6478645) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm] SKIP (https://nvbugs/6478645) From 5e40d8252068f6a76f930d9491b771f08604acc1 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 16:10:42 -0700 Subject: [PATCH 04/14] Update DSV3-Lite to retain test_auto_dtype and remove test_kv_cache_v2_nixl_python Signed-off-by: Asfiya Baig --- .../accuracy/test_disaggregated_serving.py | 83 ++++++++++--------- .../test_lists/qa/llm_function_core.txt | 7 +- .../test_lists/qa/llm_function_rtx6k.txt | 4 + .../test_lists/test-db/l0_dgx_b200.yml | 1 + .../test_lists/test-db/l0_dgx_b300.yml | 1 - .../test_lists/test-db/l0_dgx_h100.yml | 2 +- .../test_lists/test-db/l0_dgx_h200.yml | 4 + tests/integration/test_lists/waives.txt | 6 ++ 8 files changed, 63 insertions(+), 45 deletions(-) diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index 8f6a699f9c7c..b203018e7e76 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1121,35 +1121,22 @@ def test_gen_only_spec_dec(self): tensor_parallel_size=4) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) - @pytest.mark.skip_less_device(2) - @pytest.mark.skip_less_device_memory(60000) + @pytest.mark.skip_less_device(8) + @parametrize_with_ids("overlap_scheduler", [True, False]) @parametrize_with_ids("mtp_nextn", [0, 2]) - @pytest.mark.parametrize("backend", ["xgrammar", "llguidance"]) - def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): - mocker.patch.dict(os.environ, {"TRTLLM_XGUIDANCE_LENIENT": "1"}) - ctx_server_config = { - "disable_overlap_scheduler": True, - "kv_cache_config": { - "free_gpu_memory_fraction": 0.8, - }, - "guided_decoding_backend": backend, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 4096 - } + @pytest.mark.skip_less_device(8) + def test_auto_dtype(self, overlap_scheduler, mtp_nextn): + ctx_server_config = {"disable_overlap_scheduler": True} + gen_server_config = {"disable_overlap_scheduler": not overlap_scheduler} + ctx_server_config["cache_transceiver_config"] = { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 } - gen_server_config = { - "disable_overlap_scheduler": False, - "kv_cache_config": { - "free_gpu_memory_fraction": 0.8, - }, - "guided_decoding_backend": backend, - "cache_transceiver_config": { - "backend": "NIXL", - "transceiver_runtime": "PYTHON", - "max_tokens_in_buffer": 4096 - } + gen_server_config["cache_transceiver_config"] = { + "backend": "NIXL", + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 } if mtp_nextn > 0: ctx_server_config["speculative_config"] = { @@ -1171,37 +1158,51 @@ def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): } } with launch_disaggregated_llm(disaggregated_server_config, - ctx_server_config, gen_server_config, - self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["JsonModeEval"]) + ctx_server_config, + gen_server_config, + self.MODEL_PATH, + tensor_parallel_size=4) as llm: + run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) @pytest.mark.skip_less_device(2) @pytest.mark.skip_less_device_memory(60000) - @skip_pre_hopper - def test_kv_cache_v2_nixl_python(self): - """Test with use_kv_cache_manager_v2=True, block_reuse=False, backend=NIXL, transceiver_runtime=PYTHON.""" + @parametrize_with_ids("mtp_nextn", [0, 2]) + @pytest.mark.parametrize("backend", ["xgrammar", "llguidance"]) + def test_guided_decoding(self, backend: str, mtp_nextn: int, mocker): + mocker.patch.dict(os.environ, {"TRTLLM_XGUIDANCE_LENIENT": "1"}) ctx_server_config = { "disable_overlap_scheduler": True, "kv_cache_config": { - "enable_block_reuse": False, - "use_kv_cache_manager_v2": True + "free_gpu_memory_fraction": 0.8, }, + "guided_decoding_backend": backend, "cache_transceiver_config": { "backend": "NIXL", - "transceiver_runtime": "PYTHON" + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 } } gen_server_config = { - "disable_overlap_scheduler": True, + "disable_overlap_scheduler": False, "kv_cache_config": { - "enable_block_reuse": False, - "use_kv_cache_manager_v2": True + "free_gpu_memory_fraction": 0.8, }, + "guided_decoding_backend": backend, "cache_transceiver_config": { "backend": "NIXL", - "transceiver_runtime": "PYTHON" + "transceiver_runtime": "PYTHON", + "max_tokens_in_buffer": 4096 } } + if mtp_nextn > 0: + ctx_server_config["speculative_config"] = { + "decoding_type": "MTP", + "max_draft_len": mtp_nextn + } + gen_server_config["speculative_config"] = { + "decoding_type": "MTP", + "max_draft_len": mtp_nextn + } disaggregated_server_config = { "hostname": "localhost", "backend": "pytorch", @@ -1215,7 +1216,7 @@ def test_kv_cache_v2_nixl_python(self): with launch_disaggregated_llm(disaggregated_server_config, ctx_server_config, gen_server_config, self.MODEL_PATH) as llm: - run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) + run_accuracy_test(llm, self.MODEL_NAME, ["JsonModeEval"]) @pytest.mark.skip_less_device(4) @pytest.mark.skip_less_device_memory(60000) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 9faf48b249cf..5f7ec3df3841 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,11 +1,14 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync[python] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True] accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1] diff --git a/tests/integration/test_lists/qa/llm_function_rtx6k.txt b/tests/integration/test_lists/qa/llm_function_rtx6k.txt index 273f79ce5b11..4260a5328d77 100644 --- a/tests/integration/test_lists/qa/llm_function_rtx6k.txt +++ b/tests/integration/test_lists/qa/llm_function_rtx6k.txt @@ -1,3 +1,7 @@ +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml index e2d536755fdc..0b5bb8a78623 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml @@ -17,6 +17,7 @@ l0_dgx_b200: tests: - unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy - accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync # ------------- KV Cache V2 Scheduler IT (multi-GPU) --------------- - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens - kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens diff --git a/tests/integration/test_lists/test-db/l0_dgx_b300.yml b/tests/integration/test_lists/test-db/l0_dgx_b300.yml index ee04247921d4..5394ee8bc016 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b300.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b300.yml @@ -68,7 +68,6 @@ l0_dgx_b300: - accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1] - accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2] diff --git a/tests/integration/test_lists/test-db/l0_dgx_h100.yml b/tests/integration/test_lists/test-db/l0_dgx_h100.yml index 3feea471aaeb..327b7f7c8d26 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_h100.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_h100.yml @@ -34,7 +34,7 @@ l0_dgx_h100: - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False] - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend - - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_kv_cache_v2_nixl_python + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ngram - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python - accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False] diff --git a/tests/integration/test_lists/test-db/l0_dgx_h200.yml b/tests/integration/test_lists/test-db/l0_dgx_h200.yml index 2590c7f90b12..3b40b905220b 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_h200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_h200.yml @@ -27,6 +27,10 @@ l0_dgx_h200: - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8] - accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3] - accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp2] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp2] - accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=1-ctx_pp=4] diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 69a6cef9cb18..389b880060ef 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -251,6 +251,8 @@ full:GB300/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8_moe_d full:GB300/accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized SKIP (https://nvbugs/6479708) full:GB300/accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3_5_35B_A3B_VL::test_fp8_prequantized SKIP (https://nvbugs/6526429) full:GB300/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) +full:H100/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6313072) +full:H100/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6313072) full:H100/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=False] SKIP (https://nvbugs/6422343) full:H100/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=True] SKIP (https://nvbugs/6422343) full:H100/accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format SKIP (https://nvbugs/6528834) @@ -264,6 +266,8 @@ full:H100/disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_t full:H100/disaggregated/test_disaggregated.py::test_disaggregated_stress_test[input8k-output1k-conc512-qwen3_32b_fp8_stress] SKIP (https://nvbugs/6312828) full:H100/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) full:H100_PCIe/unittest/llmapi/test_llm_pytorch.py::test_llama_7b_multi_lora_evict_and_reload_lora_gpu_cache SKIP (https://nvbugs/5682551) +full:H20/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6345827) +full:H20/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6345827) full:H20/accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_fp8] SKIP (https://nvbugs/6327718) full:H20/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=False] SKIP (https://nvbugs/6422343) full:H20/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=True-v2_kv_cache=True] SKIP (https://nvbugs/6422343) @@ -297,6 +301,8 @@ full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_ full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False] SKIP (https://nvbugs/6313076) full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash SKIP (https://nvbugs/6273850) full:RTX_6000D/accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8 SKIP (https://nvbugs/6273850) +full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] SKIP (https://nvbugs/6388153) +full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] SKIP (https://nvbugs/6400067) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] SKIP (https://nvbugs/6473374) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2] SKIP (https://nvbugs/6388153) full:RTX_PRO_6000_Blackwell_Server_Edition/accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True] SKIP (https://nvbugs/6313072) From 4fc0b7d2fd292a3a34b6b0e8e57db51a3ca15e6b Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 16:36:59 -0700 Subject: [PATCH 05/14] TestDeepSeekV3Lite::test_auto_dtype_with_helix to use CPP transceiver Signed-off-by: Asfiya Baig --- tests/integration/defs/.test_durations | 1 + .../accuracy/test_disaggregated_serving.py | 85 +++++++++++++++++++ .../test_lists/qa/llm_function_core.txt | 2 + .../test_lists/test-db/l0_dgx_b200.yml | 1 + tests/integration/test_lists/waives.txt | 4 - 5 files changed, 89 insertions(+), 4 deletions(-) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 92d9aca24bd1..687b344fbc95 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -6,6 +6,7 @@ "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True]": 158.5127857142857, "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False]": 238.92535714285714, "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True]": 161.73442857142857, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2]": 276.34350980392156, "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[adp-mtp2]": 311.97010309278346, "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[noadp-mtp0]": 474.63128571428575, "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_spec_dec": 178.47664285714288, diff --git a/tests/integration/defs/accuracy/test_disaggregated_serving.py b/tests/integration/defs/accuracy/test_disaggregated_serving.py index b203018e7e76..575863d38cbb 100644 --- a/tests/integration/defs/accuracy/test_disaggregated_serving.py +++ b/tests/integration/defs/accuracy/test_disaggregated_serving.py @@ -1121,6 +1121,91 @@ def test_gen_only_spec_dec(self): tensor_parallel_size=4) as llm: run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) + @skip_pre_blackwell + @pytest.mark.skip_less_device(8) + @pytest.mark.parametrize("gen_pp,gen_tp,gen_cp,enable_attention_dp", [ + (1, 2, 2, False), + (1, 2, 2, True), + ], + ids=["pp1tp2cp2", "pp1dp2cp2"]) + @pytest.mark.parametrize("cuda_graph_config", [ + { + "enable_padding": True, + "batch_sizes": [1, 2, 4, 8, 16, 32, 64] + }, + ], + ids=["cudagraph:with_padding"]) + @pytest.mark.parametrize("comms_medium", ["fifo_v2"]) + def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config, + gen_pp, gen_tp, gen_cp, enable_attention_dp): + # Parse comms_medium to get use_nccl_for_alltoall and fifo_version. + if comms_medium == "nccl": + use_nccl_for_alltoall = True + fifo_version = 2 # Not used when NCCL is enabled. + elif comms_medium == "fifo_v1": + use_nccl_for_alltoall = False + fifo_version = 1 + elif comms_medium == "fifo_v2": + use_nccl_for_alltoall = False + fifo_version = 2 + else: + raise ValueError(f"Unknown comms_medium: {comms_medium}") + gen_ep = gen_tp * gen_cp + kv_cache_config = { + "free_gpu_memory_fraction": 0.5, + "enable_block_reuse": False, + "enable_partial_reuse": False, + "tokens_per_block": 32, + } + ctx_server_config = { + "pipeline_parallel_size": 1, + "tensor_parallel_size": 4, + "context_parallel_size": 1, + "disable_overlap_scheduler": True, + "kv_cache_config": kv_cache_config, + "enable_chunked_prefill": False, + "cuda_graph_config": None, + "cache_transceiver_config": { + "backend": "UCX", + "max_tokens_in_buffer": 8192, + }, + } + gen_server_config = { + "tensor_parallel_size": gen_tp, + "pipeline_parallel_size": gen_pp, + "context_parallel_size": gen_cp, + "moe_expert_parallel_size": gen_ep, + "cp_config": { + "cp_type": "HELIX", + "tokens_per_block": 32, + "use_nccl_for_alltoall": use_nccl_for_alltoall, + "fifo_version": fifo_version, + }, + "disable_overlap_scheduler": True, + "kv_cache_config": kv_cache_config, + "enable_chunked_prefill": False, + "cuda_graph_config": cuda_graph_config, + "cache_transceiver_config": { + "backend": "UCX", + "max_tokens_in_buffer": 8192, + }, + "enable_attention_dp": enable_attention_dp, + } + disaggregated_server_config = { + "hostname": "localhost", + "backend": "pytorch", + "context_servers": { + "num_instances": 1 + }, + "generation_servers": { + "num_instances": 1 + } + } + with launch_disaggregated_llm(disaggregated_server_config, + ctx_server_config, gen_server_config, + self.MODEL_PATH) as llm: + run_accuracy_test(llm, self.MODEL_NAME, ["MMLU", "GSM8K"]) + @pytest.mark.skip_less_device(8) @parametrize_with_ids("overlap_scheduler", [True, False]) @parametrize_with_ids("mtp_nextn", [0, 2]) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 5f7ec3df3841..a27154e48780 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,5 +1,7 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] diff --git a/tests/integration/test_lists/test-db/l0_dgx_b200.yml b/tests/integration/test_lists/test-db/l0_dgx_b200.yml index 0b5bb8a78623..f06ac5e21e6d 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_b200.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_b200.yml @@ -161,6 +161,7 @@ l0_dgx_b200: backend: pytorch orchestrator: mpi tests: + - accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] TIMEOUT (60) - accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput] TIMEOUT (60) - accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp] TIMEOUT (60) diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index 389b880060ef..b24098d34425 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -170,11 +170,7 @@ full:B200/llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_m full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/Qwen3-235B-A22B-tp16-mmlu] SKIP (https://nvbugs/6424188) full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/saved_models_Qwen3-235B-A22B_nvfp4_hf-tp16-mmlu] SKIP (https://nvbugs/6424188) full:B200/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) -<<<<<<< HEAD full:B300/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] SKIP (https://nvbugs/6410881) -full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6539942) -======= ->>>>>>> 5a1acec118 (remove TestDeepSeekV3Lite::test_auto_dtype_with_helix as helix can't run with Python transceiver) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6483370) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer] SKIP (https://nvbugs/6478645) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm] SKIP (https://nvbugs/6478645) From 903fc8be02e31a8e1090bada4a12e3387757eeb2 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 16:41:06 -0700 Subject: [PATCH 06/14] minor rearrange Signed-off-by: Asfiya Baig --- tests/integration/test_lists/qa/llm_function_core.txt | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index a27154e48780..d02c292d6413 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -1,11 +1,11 @@ accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1] accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] -accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2] +accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_sync accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0] accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2] From 6164283d86cc7272fdf8c78aad8e57d549ba0802 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 17:04:24 -0700 Subject: [PATCH 07/14] migrate deepseek configs to Python transceiver for tests/integration/defs/disaggregated/test_configs Signed-off-by: Asfiya Baig --- ...onfig_cache_aware_balance_deepseek_v3.yaml | 6 +++-- ...disagg_config_cache_reuse_deepseek_v3.yaml | 6 +++-- ...disagg_config_conditional_deepseek_v3.yaml | 6 +++-- ...config_ctxtp1_gentp1_deepseek_v3_lite.yaml | 6 +++-- ...txtp1_gentp1_deepseek_v3_lite_one_mtp.yaml | 6 +++-- ..._v3_lite_one_mtp_attention_dp_overlap.yaml | 6 +++-- ...eepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml | 6 +++-- ...txtp1_gentp1_deepseek_v3_lite_two_mtp.yaml | 6 +++-- ...config_ctxtp2_gentp2_deepseek_v3_lite.yaml | 6 +++-- ..._gentp2_deepseek_v3_lite_attention_dp.yaml | 6 +++-- ...tp2_deepseek_v3_lite_attention_dp_one.yaml | 6 +++-- ...deepseek_v3_lite_attention_dp_one_mtp.yaml | 6 +++-- ...deepseek_v3_lite_attention_dp_overlap.yaml | 6 +++-- ..._lite_attention_dp_overlap_cuda_graph.yaml | 6 +++-- ...g_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml | 2 ++ ...2_deepseek_v3_lite_overlap_cuda_graph.yaml | 6 +++-- ...ek_v3_lite_transceiver_runtime_python.yaml | 24 ------------------- ..._deepseek_v3_lite_one_mtp_block_reuse.yaml | 6 +++-- ...k_v3_lite_one_mtp_block_reuse_chunked.yaml | 6 +++-- ...ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm.yaml | 6 +++-- ...p4_gentp4_deepseek_r1_v2_fp4_tllm_mtp.yaml | 6 +++-- ...g_config_deepseek_v3_lite_empty_batch.yaml | 6 +++-- ...eepseek_v3_lite_attention_dp_gen_only.yaml | 3 ++- 23 files changed, 84 insertions(+), 65 deletions(-) delete mode 100644 tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python.yaml diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_aware_balance_deepseek_v3.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_aware_balance_deepseek_v3.yaml index 615bf8b74d41..282b9a352ff1 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_aware_balance_deepseek_v3.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_aware_balance_deepseek_v3.yaml @@ -16,7 +16,8 @@ context_servers: event_buffer_max_size: 1024 free_gpu_memory_fraction: 0.1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 2 router: @@ -29,4 +30,5 @@ generation_servers: event_buffer_max_size: 1024 free_gpu_memory_fraction: 0.1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_reuse_deepseek_v3.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_reuse_deepseek_v3.yaml index 9a51a0f5903d..3e908bcf2ab6 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_reuse_deepseek_v3.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_cache_reuse_deepseek_v3.yaml @@ -17,7 +17,8 @@ context_servers: enable_partial_reuse: true event_buffer_max_size: 1024 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 @@ -30,4 +31,5 @@ generation_servers: event_buffer_max_size: 1024 free_gpu_memory_fraction: 0.05 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_conditional_deepseek_v3.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_conditional_deepseek_v3.yaml index 7887fd2725fb..b554f840e302 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_conditional_deepseek_v3.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_conditional_deepseek_v3.yaml @@ -17,7 +17,8 @@ context_servers: event_buffer_max_size: 1024 free_gpu_memory_fraction: 0.15 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 @@ -30,4 +31,5 @@ generation_servers: event_buffer_max_size: 1024 free_gpu_memory_fraction: 0.15 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml index ce4c9b3917bf..ca673a58cf22 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml @@ -9,10 +9,12 @@ context_servers: tensor_parallel_size: 1 pipeline_parallel_size: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 pipeline_parallel_size: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp.yaml index 43081ce00a83..66bb02035726 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp.yaml @@ -13,11 +13,13 @@ context_servers: pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 pipeline_parallel_size: 1 enable_attention_dp: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_attention_dp_overlap.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_attention_dp_overlap.yaml index 77ebcc6a0585..1f33e3e7201a 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_attention_dp_overlap.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_attention_dp_overlap.yaml @@ -13,7 +13,8 @@ context_servers: enable_attention_dp: true disable_overlap_scheduler: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 @@ -21,4 +22,5 @@ generation_servers: enable_attention_dp: true disable_overlap_scheduler: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml index efbdc97c4f38..81e3c5d73baf 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_one_mtp_ctxpp2_gentp2.yaml @@ -13,7 +13,8 @@ context_servers: decoding_type: MTP max_draft_len: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 @@ -23,4 +24,5 @@ generation_servers: decoding_type: MTP max_draft_len: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml index 18cb600debcd..b69aaafc53a9 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_two_mtp.yaml @@ -13,11 +13,13 @@ context_servers: pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 pipeline_parallel_size: 1 enable_attention_dp: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite.yaml index 0d50737cc267..8fb03e4b5d7f 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite.yaml @@ -9,10 +9,12 @@ context_servers: tensor_parallel_size: 2 pipeline_parallel_size: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 pipeline_parallel_size: 1 cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp.yaml index bfec04d70572..698f2542a202 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp.yaml @@ -10,11 +10,13 @@ context_servers: pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one.yaml index 4aa309cf6cf1..3cf3225a30bc 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one.yaml @@ -10,11 +10,13 @@ context_servers: pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 pipeline_parallel_size: 1 enable_attention_dp: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one_mtp.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one_mtp.yaml index 3aa588aaa79a..1a72b6649db8 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one_mtp.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_one_mtp.yaml @@ -13,11 +13,13 @@ context_servers: pipeline_parallel_size: 1 enable_attention_dp: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 pipeline_parallel_size: 1 enable_attention_dp: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap.yaml index af8f62e920e1..fb366bdb6b71 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap.yaml @@ -10,7 +10,8 @@ context_servers: enable_attention_dp: true disable_overlap_scheduler: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 @@ -18,4 +19,5 @@ generation_servers: enable_attention_dp: true disable_overlap_scheduler: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap_cuda_graph.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap_cuda_graph.yaml index acc41bc2dc89..08e340d31cdc 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap_cuda_graph.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_attention_dp_overlap_cuda_graph.yaml @@ -9,7 +9,8 @@ context_servers: enable_attention_dp: true disable_overlap_scheduler: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 @@ -19,4 +20,5 @@ generation_servers: enable_padding: false disable_overlap_scheduler: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml index e1628021af57..aa5e722d1963 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml @@ -9,9 +9,11 @@ context_servers: pipeline_parallel_size: 1 cache_transceiver_config: backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 pipeline_parallel_size: 1 cache_transceiver_config: backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_overlap_cuda_graph.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_overlap_cuda_graph.yaml index b9d3f29b40b0..b2ed973ab018 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_overlap_cuda_graph.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_overlap_cuda_graph.yaml @@ -8,7 +8,8 @@ context_servers: pipeline_parallel_size: 1 disable_overlap_scheduler: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 2 @@ -17,4 +18,5 @@ generation_servers: enable_padding: false disable_overlap_scheduler: false cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python.yaml deleted file mode 100644 index 21856aabb69f..000000000000 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python.yaml +++ /dev/null @@ -1,24 +0,0 @@ -hostname: localhost -port: 8000 -model: DeepSeek-V3-Lite/fp8 -free_gpu_memory_fraction: 0.25 -backend: "pytorch" -disable_overlap_scheduler: True -context_servers: - num_instances: 1 - tensor_parallel_size: 2 - pipeline_parallel_size: 1 - cache_transceiver_config: - backend: "NIXL" - transceiver_runtime: "PYTHON" - urls: - - "localhost:8001" -generation_servers: - num_instances: 1 - tensor_parallel_size: 2 - pipeline_parallel_size: 1 - cache_transceiver_config: - backend: "NIXL" - transceiver_runtime: "PYTHON" - urls: - - "localhost:8002" diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse.yaml index 18eeab37fb73..8573f6e5477b 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse.yaml @@ -16,7 +16,8 @@ context_servers: kv_cache_config: enable_block_reuse: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 4 @@ -29,4 +30,5 @@ generation_servers: kv_cache_config: enable_block_reuse: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse_chunked.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse_chunked.yaml index ee58fbfd55a7..f6a6a058fe8e 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse_chunked.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp2ep2pp2_gentp4_deepseek_v3_lite_one_mtp_block_reuse_chunked.yaml @@ -18,7 +18,8 @@ context_servers: kv_cache_config: enable_block_reuse: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 4 @@ -31,4 +32,5 @@ generation_servers: kv_cache_config: enable_block_reuse: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm.yaml index 189c85a55ec0..c5c13f39c022 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm.yaml @@ -21,7 +21,8 @@ context_servers: cuda_graph_config: null print_iter_log: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON max_tokens_in_buffer: 16384 generation_servers: num_instances: 1 @@ -57,5 +58,6 @@ generation_servers: - 1024 print_iter_log: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON max_tokens_in_buffer: 16384 diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm_mtp.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm_mtp.yaml index 82902fa21f6b..724c56661b9b 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm_mtp.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp4_gentp4_deepseek_r1_v2_fp4_tllm_mtp.yaml @@ -25,7 +25,8 @@ context_servers: cuda_graph_config: null print_iter_log: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON max_tokens_in_buffer: 16384 generation_servers: num_instances: 1 @@ -64,5 +65,6 @@ generation_servers: - 1024 print_iter_log: true cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON max_tokens_in_buffer: 16384 diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_deepseek_v3_lite_empty_batch.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_deepseek_v3_lite_empty_batch.yaml index abf05da97d9d..7e2522cbf717 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_deepseek_v3_lite_empty_batch.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_deepseek_v3_lite_empty_batch.yaml @@ -17,7 +17,8 @@ context_servers: max_tokens: 512 cache_transceiver_config: max_tokens_in_buffer: 8448 - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON generation_servers: num_instances: 1 tensor_parallel_size: 1 @@ -38,6 +39,7 @@ generation_servers: max_tokens: 2560 cache_transceiver_config: max_tokens_in_buffer: 8448 - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON stream_interval: 1 num_postprocess_workers: 1 diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_gentp2_deepseek_v3_lite_attention_dp_gen_only.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_gentp2_deepseek_v3_lite_attention_dp_gen_only.yaml index 308afac71126..1ce1132e9446 100644 --- a/tests/integration/defs/disaggregated/test_configs/disagg_config_gentp2_deepseek_v3_lite_attention_dp_gen_only.yaml +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_gentp2_deepseek_v3_lite_attention_dp_gen_only.yaml @@ -14,7 +14,8 @@ generation_servers: pipeline_parallel_size: 1 enable_attention_dp: True cache_transceiver_config: - backend: DEFAULT + backend: NIXL + transceiver_runtime: PYTHON urls: - "localhost:8002" - "localhost:8003" From 0d1fdc4198b3232d8d525927db92cfcc07617dff Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Mon, 3 Aug 2026 17:07:36 -0700 Subject: [PATCH 08/14] remove disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python config from test Signed-off-by: Asfiya Baig --- tests/integration/defs/disaggregated/test_disaggregated.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/tests/integration/defs/disaggregated/test_disaggregated.py b/tests/integration/defs/disaggregated/test_disaggregated.py index c0147c8423b4..5dff46329ecd 100644 --- a/tests/integration/defs/disaggregated/test_disaggregated.py +++ b/tests/integration/defs/disaggregated/test_disaggregated.py @@ -291,8 +291,6 @@ def get_test_config(test_desc, example_dir, test_root): f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_ucx.yaml", "deepseek_v3_lite_fp8_nixl": f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml", - "deepseek_v3_lite_fp8_transceiver_runtime_python": - f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_transceiver_runtime_python.yaml", "deepseek_v3_lite_fp8_tp1": f"{test_configs_root}/disagg_config_ctxtp1_gentp1_deepseek_v3_lite.yaml", "deepseek_v3_lite_fp8_tp1_mtp": From b615d28006b782ee14d5124ff95804ea593ea26a Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 4 Aug 2026 13:50:58 -0700 Subject: [PATCH 09/14] remove test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python and associated test list entries Signed-off-by: Asfiya Baig --- .../defs/disaggregated/test_disaggregated.py | 18 ------------------ .../test_lists/qa/llm_function_core.txt | 2 -- .../test_lists/test-db/l0_dgx_h100.yml | 1 - 3 files changed, 21 deletions(-) diff --git a/tests/integration/defs/disaggregated/test_disaggregated.py b/tests/integration/defs/disaggregated/test_disaggregated.py index 5dff46329ecd..26fae1cdeef9 100644 --- a/tests/integration/defs/disaggregated/test_disaggregated.py +++ b/tests/integration/defs/disaggregated/test_disaggregated.py @@ -1902,24 +1902,6 @@ def test_disaggregated_deepseek_v3_lite_fp8_nixl(disaggregated_test_root, cwd=llm_venv.get_working_directory()) -@skip_no_hopper -@skip_arm -@pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], - indirect=True) -def test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python( - disaggregated_test_root, disaggregated_example_root, llm_venv, - deepseek_v3_model_root): - setup_model_symlink(llm_venv, deepseek_v3_model_root, - "DeepSeek-V3-Lite/fp8") - env = llm_venv._new_env.copy() - env["UCX_TLS"] = get_ucx_tls() - run_disaggregated_test(disaggregated_example_root, - "deepseek_v3_lite_fp8_transceiver_runtime_python", - env=env, - model_path=deepseek_v3_model_root, - cwd=llm_venv.get_working_directory()) - - @skip_no_hopper @skip_arm @pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index d02c292d6413..10655a8bb266 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -869,7 +869,6 @@ disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1 disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8] -disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0] @@ -944,7 +943,6 @@ disaggregated/test_auto_scaling.py::test_worker_restart[http-kv_cache_aware] disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin] disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin] disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python[TinyLlama-1.1B-Chat-v1.0] -disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_overlap_gen_first[ctx_pp1-TinyLlama-1.1B-Chat-v1.0] disaggregated/test_disaggregated.py::test_disaggregated_overlap_gen_first[ctx_pp4-TinyLlama-1.1B-Chat-v1.0] test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B] diff --git a/tests/integration/test_lists/test-db/l0_dgx_h100.yml b/tests/integration/test_lists/test-db/l0_dgx_h100.yml index 327b7f7c8d26..02983fdce158 100644 --- a/tests/integration/test_lists/test-db/l0_dgx_h100.yml +++ b/tests/integration/test_lists/test-db/l0_dgx_h100.yml @@ -214,7 +214,6 @@ l0_dgx_h100: - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_mpi[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8] - - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp_overlap[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp_one[DeepSeek-V3-Lite-fp8] From d0dbb3973cd796b60dbb13b9ca97ef6a5986bd06 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 4 Aug 2026 14:04:19 -0700 Subject: [PATCH 10/14] remove test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu and associated test list entries Signed-off-by: Asfiya Baig --- tests/integration/defs/.test_durations | 1 - .../defs/disaggregated/test_disaggregated.py | 20 ------------------- .../test_lists/qa/llm_function_core.txt | 1 - .../test_lists/test-db/l0_h100.yml | 1 - 4 files changed, 23 deletions(-) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 687b344fbc95..676a5d8d2c7b 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -637,7 +637,6 @@ "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 136.93694845360827, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 124.63738144329898, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 105.44671134020618, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 143.4193673469388, "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 73.85774285714287, "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 188.37085000000002, "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 91.19978208232446, diff --git a/tests/integration/defs/disaggregated/test_disaggregated.py b/tests/integration/defs/disaggregated/test_disaggregated.py index 26fae1cdeef9..769e3da04c0d 100644 --- a/tests/integration/defs/disaggregated/test_disaggregated.py +++ b/tests/integration/defs/disaggregated/test_disaggregated.py @@ -1902,26 +1902,6 @@ def test_disaggregated_deepseek_v3_lite_fp8_nixl(disaggregated_test_root, cwd=llm_venv.get_working_directory()) -@skip_no_hopper -@skip_arm -@pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], - indirect=True) -def test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu( - disaggregated_test_root, disaggregated_example_root, llm_venv, - deepseek_v3_model_root): - setup_model_symlink(llm_venv, deepseek_v3_model_root, - "DeepSeek-V3-Lite/fp8") - env = llm_venv._new_env.copy() - env["TRTLLM_USE_UCX_KVCACHE"] = "1" - env["UCX_TLS"] = get_ucx_tls() - - run_disaggregated_test(disaggregated_example_root, - "deepseek_v3_lite_fp8_tp1", - env=env, - model_path=deepseek_v3_model_root, - cwd=llm_venv.get_working_directory()) - - @skip_no_hopper @pytest.mark.skip_less_device(4) @pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index 10655a8bb266..d96cf043924b 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -870,7 +870,6 @@ disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1 disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8] -disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0] disaggregated/test_disaggregated.py::test_disaggregated_genbs1[TinyLlama-1.1B-Chat-v1.0] disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b] diff --git a/tests/integration/test_lists/test-db/l0_h100.yml b/tests/integration/test_lists/test-db/l0_h100.yml index 0c15bf0e61b7..021ab0898c07 100644 --- a/tests/integration/test_lists/test-db/l0_h100.yml +++ b/tests/integration/test_lists/test-db/l0_h100.yml @@ -171,7 +171,6 @@ l0_h100: - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8] - - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0] - disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0] - disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8] From 6408eafc8bcc32e1f1c758f925b645297fa2fed8 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 4 Aug 2026 14:10:34 -0700 Subject: [PATCH 11/14] Revert "remove test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu and associated test list entries" This reverts commit d1eb9965c85a31a1cbecd4a378f16710c2906ef1. --- tests/integration/defs/.test_durations | 1 + .../defs/disaggregated/test_disaggregated.py | 20 +++++++++++++++++++ .../test_lists/qa/llm_function_core.txt | 1 + .../test_lists/test-db/l0_h100.yml | 1 + 4 files changed, 23 insertions(+) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 676a5d8d2c7b..687b344fbc95 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -637,6 +637,7 @@ "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 136.93694845360827, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 124.63738144329898, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 105.44671134020618, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 143.4193673469388, "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 73.85774285714287, "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 188.37085000000002, "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 91.19978208232446, diff --git a/tests/integration/defs/disaggregated/test_disaggregated.py b/tests/integration/defs/disaggregated/test_disaggregated.py index 769e3da04c0d..26fae1cdeef9 100644 --- a/tests/integration/defs/disaggregated/test_disaggregated.py +++ b/tests/integration/defs/disaggregated/test_disaggregated.py @@ -1902,6 +1902,26 @@ def test_disaggregated_deepseek_v3_lite_fp8_nixl(disaggregated_test_root, cwd=llm_venv.get_working_directory()) +@skip_no_hopper +@skip_arm +@pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], + indirect=True) +def test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu( + disaggregated_test_root, disaggregated_example_root, llm_venv, + deepseek_v3_model_root): + setup_model_symlink(llm_venv, deepseek_v3_model_root, + "DeepSeek-V3-Lite/fp8") + env = llm_venv._new_env.copy() + env["TRTLLM_USE_UCX_KVCACHE"] = "1" + env["UCX_TLS"] = get_ucx_tls() + + run_disaggregated_test(disaggregated_example_root, + "deepseek_v3_lite_fp8_tp1", + env=env, + model_path=deepseek_v3_model_root, + cwd=llm_venv.get_working_directory()) + + @skip_no_hopper @pytest.mark.skip_less_device(4) @pytest.mark.parametrize("deepseek_v3_model_root", ['DeepSeek-V3-Lite-fp8'], diff --git a/tests/integration/test_lists/qa/llm_function_core.txt b/tests/integration/test_lists/qa/llm_function_core.txt index d96cf043924b..10655a8bb266 100644 --- a/tests/integration/test_lists/qa/llm_function_core.txt +++ b/tests/integration/test_lists/qa/llm_function_core.txt @@ -870,6 +870,7 @@ disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1 disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8] +disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8] disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0] disaggregated/test_disaggregated.py::test_disaggregated_genbs1[TinyLlama-1.1B-Chat-v1.0] disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b] diff --git a/tests/integration/test_lists/test-db/l0_h100.yml b/tests/integration/test_lists/test-db/l0_h100.yml index 021ab0898c07..0c15bf0e61b7 100644 --- a/tests/integration/test_lists/test-db/l0_h100.yml +++ b/tests/integration/test_lists/test-db/l0_h100.yml @@ -171,6 +171,7 @@ l0_h100: - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_single_gpu_mtp[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_tp1_two_mtp[DeepSeek-V3-Lite-fp8] + - disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8] - disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0] - disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0] - disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8] From 5536469e16437dccc8ee420eadfc7830e41d31f7 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Tue, 4 Aug 2026 14:15:14 -0700 Subject: [PATCH 12/14] Add config for TP1 UCX test Signed-off-by: Asfiya Baig --- ...nfig_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml | 17 +++++++++++++++++ .../defs/disaggregated/test_disaggregated.py | 8 +++++--- 2 files changed, 22 insertions(+), 3 deletions(-) create mode 100644 tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml diff --git a/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml new file mode 100644 index 000000000000..2ee6d65e5e0d --- /dev/null +++ b/tests/integration/defs/disaggregated/test_configs/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml @@ -0,0 +1,17 @@ +hostname: localhost +model: DeepSeek-V3-Lite/fp8 +free_gpu_memory_fraction: 0.25 +backend: pytorch +disable_overlap_scheduler: true +context_servers: + num_instances: 1 + tensor_parallel_size: 1 + pipeline_parallel_size: 1 + cache_transceiver_config: + backend: UCX +generation_servers: + num_instances: 1 + tensor_parallel_size: 1 + pipeline_parallel_size: 1 + cache_transceiver_config: + backend: UCX diff --git a/tests/integration/defs/disaggregated/test_disaggregated.py b/tests/integration/defs/disaggregated/test_disaggregated.py index 26fae1cdeef9..ab14ed89c8c4 100644 --- a/tests/integration/defs/disaggregated/test_disaggregated.py +++ b/tests/integration/defs/disaggregated/test_disaggregated.py @@ -287,7 +287,9 @@ def get_test_config(test_desc, example_dir, test_root): f"{test_configs_root}/disagg_config_ctxpp4_gentp4.yaml", "deepseek_v3_lite_fp8_mpi": f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_mpi.yaml", - "deepseek_v3_lite_fp8_ucx": + "deepseek_v3_lite_fp8_tp1_ucx": + f"{test_configs_root}/disagg_config_ctxtp1_gentp1_deepseek_v3_lite_ucx.yaml", + "deepseek_v3_lite_fp8_tp2_ucx": f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_ucx.yaml", "deepseek_v3_lite_fp8_nixl": f"{test_configs_root}/disagg_config_ctxtp2_gentp2_deepseek_v3_lite_nixl.yaml", @@ -1874,7 +1876,7 @@ def test_disaggregated_deepseek_v3_lite_fp8_ucx(disaggregated_test_root, env["TRTLLM_USE_UCX_KVCACHE"] = "1" env["UCX_TLS"] = get_ucx_tls() run_disaggregated_test(disaggregated_example_root, - "deepseek_v3_lite_fp8_ucx", + "deepseek_v3_lite_fp8_tp2_ucx", env=env, model_path=deepseek_v3_model_root, cwd=llm_venv.get_working_directory()) @@ -1916,7 +1918,7 @@ def test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu( env["UCX_TLS"] = get_ucx_tls() run_disaggregated_test(disaggregated_example_root, - "deepseek_v3_lite_fp8_tp1", + "deepseek_v3_lite_fp8_tp1_ucx", env=env, model_path=deepseek_v3_model_root, cwd=llm_venv.get_working_directory()) From 569c7c53b264079f97f6650794ceddeec1076fcb Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Wed, 5 Aug 2026 15:37:32 -0700 Subject: [PATCH 13/14] fix rebase bugs Signed-off-by: Asfiya Baig --- tests/integration/defs/.test_durations | 3070 ++++++++++++----------- tests/integration/test_lists/waives.txt | 1 + 2 files changed, 1570 insertions(+), 1501 deletions(-) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 687b344fbc95..6d7ec789d334 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -1,1503 +1,1571 @@ { - "accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 566.384125, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False]": 605.8393098591549, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 792.857375, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False]": 262.5967857142857, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True]": 158.5127857142857, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False]": 238.92535714285714, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True]": 161.73442857142857, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2]": 276.34350980392156, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[adp-mtp2]": 311.97010309278346, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[noadp-mtp0]": 474.63128571428575, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_spec_dec": 178.47664285714288, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0]": 409.83045454545453, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2]": 744.224909090909, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 205.94233333333335, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 188.42866379310345, - "accuracy/test_disaggregated_serving.py::TestDeepSeekV4Flash::test_auto_dtype": 592.6489204545454, - "accuracy/test_disaggregated_serving.py::TestGLM52NVFP4::test_nvfp4_nixl_python[cache_mgr_v1]": 990.9347659574468, - "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 342.7804705882353, - "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 415.1615625, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False]": 245.6970612244898, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True]": 247.0623829787234, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 255.7429157894737, - "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 424.93321978021976, - "accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4": 1390.96909375, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-False]": 208.08547872340426, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-True]": 209.256806122449, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-False]": 216.41998947368418, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-True]": 216.31725263157895, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-False]": 166.9120744680851, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-True]": 164.41655319148936, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-False]": 179.65035714285713, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-True]": 175.9757340425532, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-False]": 204.8307142857143, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-True]": 205.07693617021278, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-False]": 216.36536082474225, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-True]": 217.35315463917524, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-False]": 166.50253061224493, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-True]": 164.4506326530612, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-False]": 180.33906382978722, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-True]": 177.16786734693878, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_beam_search": 819.1626363636364, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_eagle3[eagle3_one_model=True-overlap_scheduler=True]": 171.46033653846152, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_gen_only_spec_dec": 217.6168762886598, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 76.94890816326532, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 156.0001489361702, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python": 520.4379043478261, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[GSM8K]": 175.96124, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[MMLU]": 102.43104, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp1]": 143.1969090909091, - "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp1]": 90.42032352941176, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=False]": 391.84766666666667, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=3-block_reuse=True-use_py_transceiver=False]": 478.8976666666667, - "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_ctx_dp2_gen_tp4": 281.545, - "accuracy/test_disaggregated_serving.py::TestQwen3NextInstruct::test_auto_dtype[use_py_transceiver=True]": 432.71708, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False]": 347.91988297872336, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-True]": 340.39185714285713, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[True-True]": 329.44937234042555, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2]": 326.08301020408163, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill": 348.02151020408166, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first": 1046.73604, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first_kv_cache_v1": 124.06744444444443, - "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend": 188.3538899082569, - "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_nvfp4]": 332.06218233618233, - "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[qwen3vl_2b_instruct]": 185.360093220339, - "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[no_overlap]": 69.15029714285714, - "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[overlap]": 67.1641367781155, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-ep2]": 177.65254545454545, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-tp2]": 293.10047959183674, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b]": 148.66536363636365, - "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[20b]": 130.28418181818182, - "accuracy/test_llm_api_autodeploy.py::TestGemma4MoE::test_bf16": 1040.9451304347826, - "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma3n_e2b_it": 652.3853028571428, - "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma4_e2b_it": 345.52021037463976, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_attention_dp[4]": 281.7136451612903, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[torch-True-1]": 909.5608181818183, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[triton-False-1]": 320.45450417827294, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-1]": 152.23809254143646, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-4]": 138.5165392670157, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-True-1]": 328.91990909090913, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[flashinfer]": 207.7224, - "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[trtllm]": 184.89575555555555, - "accuracy/test_llm_api_autodeploy.py::TestMiniMaxM2::test_finegrained_fp8": 607.0751460674157, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[Qwen_QwQ-32B-False]": 262.3381111111111, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[deepseek-ai_DeepSeek-R1-0528-True]": 809.233, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[google_gemma-3-1b-it-False]": 61.70166666666666, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.1-8B-Instruct-False]": 50.529333333333334, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.3-70B-Instruct-False]": 145.9032222222222, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Codestral-22B-v0.1-False]": 90.96744444444444, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Ministral-8B-Instruct-2410-False]": 64.63533333333334, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_DeepSeek-R1-0528-NVFP4-v2-True]": 1319.9854545454546, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-FP8-True]": 110.99866666666667, - "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-NVFP4-True]": 125.70889795918367, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-flashinfer_ssm-False]": 495.21763636363636, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-False]": 349.2198181818182, - "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-True]": 464.39036363636365, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-1-attn_dp_off-trtllm]": 1240.6362727272726, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 352.4914444444444, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-1-attn_dp_off-trtllm]": 393.3906777316736, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 259.3917204301075, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 219.00550537634408, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 315.08336363636363, - "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-4-attn_dp_off-trtllm]": 305.414, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 603.2348181818182, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 466.7068367346939, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 405.1242727272727, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 725.3176630727763, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-4-attn_dp_on-trtllm]": 510.6742727272727, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[bf16]": 240.1639674796748, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[fp8]": 225.47681818181817, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer]": 485.037, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm]": 375.2629090909091, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[fp8_ws4_80gb-trtllm]": 506.8681808510638, - "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[nvfp4_ws4_80gb-trtllm]": 512.3643636363637, - "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-4]": 1732.888907216495, - "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-8]": 2127.4628181818184, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[False]": 400.33763636363636, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[True]": 503.32218181818183, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_fp8[True]": 515.5300969101123, - "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_nvfp4[True]": 417.8357272727273, - "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_bf16_small[4]": 499.28675714285714, - "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_nvfp4[8]": 1178.8152727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[latency]": 1994.9886428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput]": 696.7412417218543, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency]": 815.1006, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp]": 445.61188888888887, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp_tp4]": 752.4664851485148, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen]": 924.8983000000001, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen_adp_lmtp]": 459.0964, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput]": 455.72643853820597, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_bs8_mtp]": 636.5326666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp]": 469.2359897260274, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_pp4_mtp]": 943.4426666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp4]": 888.5035, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8]": 409.552, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_dsa_host_cache_offload[host_cache_offload]": 960.41075, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline]": 899.9911666666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline_mtp1]": 1043.9808039215686, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_attn_multi_gpus": 693.5356666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline]": 435.158900990099, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline_mtp1]": 384.64672277227726, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[baseline]": 424.88255208333334, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3::test_skip_softmax_attention_multi_gpus[target_sparsity_0.9]": 1936.0571020408163, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 109.189, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 111.4233, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 131.53244444444445, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 131.10222222222222, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 103.06677777777779, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 107.91711111111111, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 111.58439999999999, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 117.0749, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 147.24383333333336, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 150.32771428571428, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 106.54339999999999, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 109.9482, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 129.39533333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 176.43553470437018, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 250.69174673629243, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 105.5962, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 111.16025, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 120.65477777777778, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 130.57385714285715, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 100.0108, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 107.22133333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 185.2611, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 185.54422222222223, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 104.255, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 109.0158888888889, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 243.6549696969697, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=False]": 423.52147860962566, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=True]": 681.503816, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp": 216.84545161290322, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 75.69972727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 166.08079999999998, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 95.51, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 54.731818181818184, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.107, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 51.38036363636364, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.14818181818182, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 43.47654545454546, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 131.0839090909091, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 37.365199999999994, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 260.33008163265305, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[pp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 308.887675, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 54.127272727272725, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 47.00145454545454, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 44.096, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 48.436090909090915, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 50.10754545454546, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 50.03754545454546, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 61.07272727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 63.22081818181818, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 249.28828947368422, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 64.609, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 173.59127272727272, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 45.734, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 44.64572727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 54.545, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.67136363636364, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 69.7822, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 44.198, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 69.9719090909091, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 235.08795999999998, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 39.305, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 67.627, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=0]": 94.18960000000001, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=2]": 99.85809090909092, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=0]": 57.95333333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=2]": 70.299, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=0]": 72.82077777777778, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=2]": 74.54666666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=False]": 196.24513333333334, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=True]": 240.260800807537, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 257.00025, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=True]": 2899.535428571429, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 242.9692307692308, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 1214.0928999999999, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 279.8845, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=True]": 3115.894142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 339.368, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 819.28775, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=fp8-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 435.4769372496662, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]": 214.80394305555555, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 227.8988691983122, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 294.3789699453552, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 279.15114520547945, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_dummy_load_format": 169.91534423897582, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 127.89110000000001, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 123.1895, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 106.7188, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 122.71088888888889, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 109.73866666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 122.89022222222222, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 104.65889999999999, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 201.32696816208394, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 107.83333333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 103.43469999999999, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 99.9161111111111, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 126.5578888888889, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 102.3408, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 105.42333333333333, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 205.4267492063492, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 107.1272, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 127.02718181818183, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.298272727272725, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 48.119181818181815, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 129.86932352941176, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 72.26036363636364, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 46.696, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 73.61151960784314, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 65.76149019607843, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 98.89809090909091, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.159454545454544, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 66.94454545454546, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 51.37372727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 49.74481818181818, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 49.787454545454544, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 74.08118181818182, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 66.64172727272728, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 76.389, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 59.23609090909091, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 58.95272727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.89263636363636, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-sampler_async_worker=False]": 80.8312, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.64163636363636, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 53.18277777777778, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 134.4777777777778, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 67.16441176470589, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 61.92367647058824, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 113.32422222222222, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=True]": 115.1615925925926, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 51.46609090909091, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 66.509, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.761, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.95254545454546, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 134.4498148148148, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=0]": 68.2290737704918, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=2]": 154.16124725274724, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=0]": 51.68824509803922, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=2]": 60.39678217821782, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 79.26566666666668, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 223.61991005291003, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]": 110.20872727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=2]": 64.23452941176471, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=fp8-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 98.88074127906977, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 238.25687696335078, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 193.4495, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 325.0439166666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 110.1784, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 130.3401, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 108.9846, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.9392, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 119.118375, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 138.20675, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 117.565, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 158.50493922651935, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 121.72740331491713, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 172.87279999999998, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 171.1799, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 314.79475447570337, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.463, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 108.3047850340136, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 208.770375, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 201.36658333333335, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 189.09936363636365, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 417.9765, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 196.3797142857143, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 236.29854545454543, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 119.75618181818183, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-pp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 246.47781818181818, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 155.82318181818184, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 382.13807936507936, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 164.28670588235295, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 140.84088888888888, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 55.83880952380952, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 109.41854545454545, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 229.1358, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 99.4054, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 131.42146153846153, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 123.13945454545454, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 114.74166666666667, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 82.9781914893617, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 147.91820975609755, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 161.596, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 146.64896595744682, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 140.86234782608693, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 150.08861904761906, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 144.58642424242424, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 132.33410526315788, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 133.45738095238096, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 73.9634, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 127.70872727272727, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 177.698671875, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 73.15577570093458, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 211.7968248847926, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 203.5012, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=True-moe_backend=TRTLLM]": 132.16976041666666, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_batch_waiting[batch_wait_timeout_iters=10-batch_wait_max_tokens_ratio=1.0-mtp_nextn=0-fp8kv=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-v2_kv_cache=True]": 208.58932602739728, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=MEGAMOE_DEEPGEMM]": 737.5082040816327, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=TRTLLM]": 366.2747959183673, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy": 1131.2820365853659, - "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4ProDSpark::test_gsm8k_dep8_megamoe_deepgemm": 896.7543000000001, - "accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4[tp_size=8-ep_size=8]": 694.6597543859649, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dflash": 587.7711835616439, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format": 87.91589784946235, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[cutlass-one_model-overlap_scheduler]": 1955.2867352941175, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[triton-one_model-overlap_scheduler]": 1010.3446470588235, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-no_overlap_scheduler]": 688.9055999999999, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-overlap_scheduler]": 439.4791601941747, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-no_overlap_scheduler]": 426.94371428571424, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-overlap_scheduler]": 444.78338647342997, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-no_overlap_scheduler]": 550.945875, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-overlap_scheduler]": 630.1294747474748, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-no_overlap_scheduler]": 706.0050833333333, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-overlap_scheduler]": 591.5004950980392, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-one_model]": 280.22, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-two_model]": 398.28516666666667, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v2_kv_cache-one_model]": 1526.7824285714287, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-cutlass-auto]": 273.2275148648649, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-auto]": 241.6941, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-fp8]": 207.31851506849316, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-cutlass-auto]": 175.9200138121547, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-auto]": 248.3882, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-fp8]": 275.5822980501393, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-triton-auto]": 1710.5607717391304, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-auto]": 587.4411111111111, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-fp8]": 455.14378974358976, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-triton-auto]": 549.2571739130434, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-trtllm-auto]": 453.9397747747748, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-triton-auto]": 463.44741304347826, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-trtllm-auto]": 552.508797752809, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-triton-auto]": 900.5855666666666, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-auto]": 586.8257777777777, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-fp8]": 654.6051242937854, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-triton-auto]": 646.6367173913043, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-auto]": 508.7110540540541, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-fp8]": 469.1895714285714, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-tp4-trtllm-auto]": 634.4464065934066, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache_no_reuse-tp4-cutlass-auto]": 370.1528152173913, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus_online_eplb[fp8]": 282.8789893617021, - "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]": 327.9946086956522, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype": 201.60288607594936, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_reuse": 115.71518618618619, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_without_reuse": 128.66493072289157, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse": 49.06097674418604, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_disable_overlap_scheduler": 51.77953230769231, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_no_partial_reuse": 57.0330583090379, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_partial_reuse": 60.72586086956522, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_partial_reuse": 49.97132530120482, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse": 43.764580281690144, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_disable_overlap_scheduler": 47.694558641975306, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_low_memory_available": 57.55918452380953, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=False]": 183.97511111111112, - "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=True]": 188.8469, - "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_auto_dtype": 933.2313, - "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_fp8_prequantized": 419.3982222222223, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8]": 467.342, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8]": 657.7212, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8]": 1009.5471894736842, - "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8_attn_dp]": 682.5311111111112, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_bf16": 565.337822622108, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_fp8": 313.896358649789, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_nvfp4": 268.91661690885076, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_bf16_dflash": 331.3741212121212, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_fp8_dflash": 313.25575872093026, - "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_nvfp4_dflash": 356.3046358543417, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4": 148.77345046235138, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_4]": 37.77284679665738, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_64]": 136.69611111111112, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=False]": 87.9016, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=True]": 140.07416666666666, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=False]": 82.0882, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=True]": 114.42385714285714, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=FLASHINFER-torch_compile=False]": 134.89218181818183, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=TRTLLM-torch_compile=False]": 124.3952, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=FLASHINFER-torch_compile=False]": 46.56345454545455, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=False]": 41.564454545454545, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=True]": 46.043454545454544, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=False]": 86.29798181818181, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=True]": 68.56609090909092, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=TRTLLM-torch_compile=True]": 121.82825, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=FLASHINFER]": 157.36079999999998, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=TRTLLM]": 48.425628985507245, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=FLASHINFER]": 170.8326, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=TRTLLM]": 75.20133526011561, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dflash": 147.7249836065574, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dummy_load_format": 17.48164864864865, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=False-eagle3_one_model=True-overlap_scheduler=True]": 998.3113925501433, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=True-eagle3_one_model=True-overlap_scheduler=True]": 1011.6907043478261, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 74.59066666666668, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 110.4135, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 54.13452586206896, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 81.929431085044, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 73.4508888888889, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 110.1177, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 49.26115269461078, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 70.60236734693878, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[pp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 90.87263636363637, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 36.40436363636364, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 37.01018181818181, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 105.15461111111111, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 74.54427272727273, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 126.09972906403941, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 90.54714427860696, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 39.76645454545454, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 145.214, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 58.84855392156863, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 80.60036, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[llguidance]": 47.924777777777784, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 48.18807374631269, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[llguidance]": 47.427454545454545, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[llguidance-eagle3_one_model=True]": 58.863800000000005, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 51.88051724137931, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[llguidance]": 47.401900000000005, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[xgrammar]": 47.24155555555556, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=False]": 57.35074229691877, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=True]": 97.46340384615384, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=True-attn_backend=TRTLLM-torch_compile=True]": 179.64886426592798, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=False]": 785.3822138728324, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=True]": 745.1532246153846, - "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B_Instruct_RocketKV::test_auto_dtype": 924.9611, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 592.7475376344086, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=True]": 656.0788791208792, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 687.6400107526881, - "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=True]": 763.7198031914893, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_auto_dtype[tp_size=8-ep_size=8]": 10800.0, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_mxfp8[use_msa=False]": 10800.0, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_mxfp8_piecewise_cuda_graph[use_msa=False]": 10800.0, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=False]": 10800.0, - "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=True]": 1042.919387755102, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_fp8": 281.42, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_nvfp4_marlin_multi_gpus[tp_size=8]": 354.6746428571428, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-cpp_mamba_cache]": 373.539, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-python_mamba_cache]": 320.75507692307696, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-cpp_mamba_cache]": 214.66808333333336, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-python_mamba_cache]": 220.42576923076922, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar": 305.8700747663552, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar_custom_op": 212.21884615384616, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_OFF]": 408.7718888888889, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_ON]": 368.3760617977528, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[TEP4]": 630.2835833333334, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_hopper_w4a16": 509.27528571428576, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_online_eplb[moe_backend=TRTLLM]": 381.3987, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 341.6574444444444, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp": 666.2585544554455, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp_custom_op": 386.734, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3]": 838.8766666666667, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8]": 464.279, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_block_reuse[ADP4_MTP]": 1030.6662569832401, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 867.7993636363636, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_static_eplb[moe_backend=CUTLASS]": 470.4553846153846, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_off-trtllm]": 552.3911145833333, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_on-cutedsl]": 657.688081632653, - "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus": 2768.8230714285714, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_2gpu_mtp_ar": 198.87144444444445, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[dep4]": 341.21818181818185, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-cutlass]": 205.73988888888888, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-trtllm]": 178.89322222222222, - "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[tp1_block_reuse-cutlass]": 472.1311, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]": 230.0092222222222, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp]": 297.94870000000003, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_cutlass]": 1410.4496153846155, - "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_trtllm_eagle3]": 326.27490476190474, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_dummy_load_format": 45.34000611620795, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=False]": 151.40441666666666, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=True]": 188.0913354037267, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=False]": 127.05255555555556, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=True]": 157.40188888888886, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]": 138.786, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=True]": 176.966, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=False]": 137.82438095238095, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=True]": 125.6016, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=False]": 304.94709523809524, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=True]": 222.94661904761907, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=False]": 197.08683333333335, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=True]": 177.8886295264624, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=False]": 350.6162083333333, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=True]": 290.07529411764705, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=False]": 188.4275, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]": 222.8463482142857, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a16_mxfp4[latency-TRTLLM]": 99.26474238227146, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-CUTLASS]": 90.6996, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-TRTLLM]": 104.65530578512397, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-CUTLASS]": 85.41755555555557, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-TRTLLM]": 77.00845762711864, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention[target_sparsity_0.9-fp8kv=True]": 845.3932931506849, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=False]": 704.4539090909091, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=True]": 539.9865714285714, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False]": 615.3447894736842, - "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=True]": 547.9886875, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-CUTLASS]": 213.385125, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-TRTLLM]": 208.752, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-CUTLASS]": 225.09233333333333, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM]": 264.60998076923073, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_mtp": 355.00717987804876, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False]": 418.67988888888885, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_cutedsl]": 558.5049900990099, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_trtllm]": 438.426, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_block_reuse]": 627.075405940594, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_trtllm]": 475.7063333333333, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 948.6269726775956, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 569.6396262626263, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_mtp3_gdn_replay_tep4": 529.1704337349398, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_bf16": 188.67048780487804, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash": 260.3592507374631, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8": 190.37416863905324, - "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8_piecewise_cuda_graph": 267.21039432176656, - "accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM]": 225.70839265536725, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dflash": 166.4091335403727, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dummy_load_format": 39.61743076923077, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=False]": 100.84025382262996, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=True]": 118.26863636363636, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=True-enable_draft_len_schedule=False]": 113.90258333333333, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=True-enable_max_concurrency=False-enable_draft_len_schedule=False]": 198.20316666666665, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales[latency]": 205.6853557623479, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales_early_first_token_response": 116.59400867052022, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[fp8-latency]": 37.92944324324324, - "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[mxfp8-latency]": 53.90125, - "accuracy/test_llm_api_pytorch.py::TestSeedOss_36B::test_auto_dtype": 1173.8134285714286, - "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_auto_dtype[tp_size=8-ep_size=8]": 5400.0, - "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_fp8_block_scales[tp_size=4-ep_size=4-mtp_nextn=0]": 5400.0, - "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_fp8_block_scales[tp_size=4-ep_size=4-mtp_nextn=3]": 5400.0, - "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_nvfp4[tp_size=4-ep_size=4-mtp_nextn=0]": 5400.0, - "accuracy/test_llm_api_pytorch.py::TestStep3_7::test_nvfp4[tp_size=4-ep_size=4-mtp_nextn=3]": 5400.0, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_cuda_graph_matches_eager_logits[tinyllama-1.1b]": 14.05920506329114, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[gemma-3-1b]": 19.97921739130435, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[phi-4-mini]": 22.435924242424242, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen2-7b]": 24.917961636828643, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen3-0.6b]": 9.642792307692307, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[starcoder2-3b]": 33.13732070707071, - "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[tinyllama-1.1b]": 6.582192405063291, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_cuda_graph_matches_eager_logits[bert-yelp]": 6.682479797979798, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-cuda_graph]": 4.394005050505051, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-eager]": 4.2008080808080805, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_per_token_reward[qwen2.5-prm-7b]": 22.808949494949495, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-0.6b]": 12.521669789227166, - "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-8b]": 33.73845667447307, - "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized": 325.59247987616095, - "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma4_26B_A4B::test_nvfp4": 374.86357142857145, - "accuracy/test_llm_api_pytorch_multimodal.py::TestKimiK25::test_nvfp4[dep8]": 2679.2065, - "accuracy/test_llm_api_pytorch_multimodal.py::TestMistralSmall24B::test_auto_dtype[forced_chunked_prefill]": 181.28377777777777, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[bf16]": 400.345, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8]": 690.9473907692308, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8_mmmu_encoder_cuda_graph]": 346.34280115273776, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]": 416.1755833333333, - "accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype[forced_chunked_prefill]": 286.4858888888889, - "accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL::test_auto_dtype[forced_chunked_prefill]": 177.8486, - "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_fp8_block_scales[mtp_nextn=3]": 7200.0, - "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_nvfp4[mtp_nextn=0]": 7200.0, - "accuracy/test_llm_api_pytorch_multimodal.py::TestStep3_7::test_nvfp4[mtp_nextn=3]": 7200.0, - "cpp/test_multi_gpu.py::test_cache_transceiver[2proc-ucx_kvcache-90]": 178.4007340425532, - "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-nixl_kvcache-90]": 242.4074680851064, - "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-ucx_kvcache-90]": 275.1460652173913, - "cpp/test_multi_gpu.py::test_fused_gemm_allreduce[4proc-90]": 21.4115, - "cpp/test_multi_gpu.py::test_mpi_utils[90]": 220.0863829787234, - "cpp/test_multi_gpu.py::test_user_buffer[2proc-90]": 4.284468085106383, - "cpp/test_unit_tests.py::test_unit_tests[batch_manager-80]": 552.2707548209366, - "cpp/test_unit_tests.py::test_unit_tests[common-80]": 223.79825, - "cpp/test_unit_tests.py::test_unit_tests[common-90]": 222.75193593314765, - "cpp/test_unit_tests.py::test_unit_tests[executor-80]": 340.2329166666667, - "cpp/test_unit_tests.py::test_unit_tests[kernels-80]": 789.6245833333334, - "cpp/test_unit_tests.py::test_unit_tests[kernels-90]": 768.3216211699165, - "cpp/test_unit_tests.py::test_unit_tests[layers-80]": 913.9700833333334, - "cpp/test_unit_tests.py::test_unit_tests[layers-90]": 805.0914456824512, - "cpp/test_unit_tests.py::test_unit_tests[runtime-80]": 689.6249166666666, - "cpp/test_unit_tests.py::test_unit_tests[thop-80]": 3.291, - "cpp/test_unit_tests.py::test_unit_tests[thop-90]": 3.207640668523677, - "disaggregated/test_ad_disagg.py::test_async_eagle3_full_model_handoff": 97.89396703296703, - "disaggregated/test_ad_disagg.py::test_async_generation_matches_aggregate": 88.92743956043955, - "disaggregated/test_ad_disagg.py::test_async_generation_no_overlap_matches_aggregate": 87.01123076923076, - "disaggregated/test_ad_disagg.py::test_async_sharded_generation_handoff": 99.30441758241757, - "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[deepseek_v3_mla]": 161.1555747126437, - "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[tinyllama]": 98.48647550432277, - "disaggregated/test_ad_disagg.py::test_disaggregated_logits[deepseek_v3_mla]": 58.13549287749288, - "disaggregated/test_ad_disagg.py::test_disaggregated_logits[tinyllama]": 36.24735243553008, - "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[deepseek_v3_mla]": 142.15129971181557, - "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[tinyllama]": 97.21402017291066, - "disaggregated/test_ad_disagg.py::test_tinyllama_batch_handoff_semantic_slots": 77.39993123209169, - "disaggregated/test_ad_disagg_trtllm_serve.py::test_openai_completion": 62.068666666666665, - "disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin]": 107.51943, - "disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin]": 136.50441836734694, - "disaggregated/test_auto_scaling.py::test_minimal_instances[etcd-round_robin]": 105.13314285714286, - "disaggregated/test_auto_scaling.py::test_minimal_instances[http-round_robin]": 103.02789690721649, - "disaggregated/test_auto_scaling.py::test_service_discovery[etcd-round_robin]": 54.86334020618557, - "disaggregated/test_auto_scaling.py::test_service_discovery[http-kv_cache_aware]": 52.12690909090909, - "disaggregated/test_auto_scaling.py::test_service_discovery[http-round_robin]": 52.24916831683168, - "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-load_balancing]": 210.1801530612245, - "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-round_robin]": 207.23043564356436, - "disaggregated/test_auto_scaling.py::test_worker_restart[http-load_balancing]": 203.67193, - "disaggregated/test_disaggregated.py::test_disaggregated_benchmark_gen_only_insufficient_kv[TinyLlama-1.1B-Chat-v1.0]": 74.8770512195122, - "disaggregated/test_disaggregated.py::test_disaggregated_cache_aware_balance[TinyLlama-1.1B-Chat-v1.0]": 126.79705569007264, - "disaggregated/test_disaggregated.py::test_disaggregated_chat_completion_tool_calls[TinyLlama-1.1B-Chat-v1.0]": 70.9736107055961, - "disaggregated/test_disaggregated.py::test_disaggregated_conditional[TinyLlama-1.1B-Chat-v1.0]": 91.700578313253, - "disaggregated/test_disaggregated.py::test_disaggregated_cuda_graph[TinyLlama-1.1B-Chat-v1.0]": 81.07452153110047, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_conditional_v2[DeepSeek-V3-Lite-bf16]": 144.77602127659574, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_empty_batch[DeepSeek-V3-Lite-bf16]": 195.28455670103094, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 136.93694845360827, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 124.63738144329898, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 105.44671134020618, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 143.4193673469388, - "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 73.85774285714287, - "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 188.37085000000002, - "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 91.19978208232446, - "disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0]": 83.23244171779142, - "disaggregated/test_disaggregated.py::test_disaggregated_mixed[TinyLlama-1.1B-Chat-v1.0]": 131.9174964028777, - "disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_test[req120-conc64-qwen3_32b_fp8_mixed_stress]": 220.23792307692307, - "disaggregated/test_disaggregated.py::test_disaggregated_multi_gpu[TinyLlama-1.1B-Chat-v1.0]": 73.60878787878788, - "disaggregated/test_disaggregated.py::test_disaggregated_ngram[TinyLlama-1.1B-Chat-v1.0]": 87.51331400966183, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap[TinyLlama-1.1B-Chat-v1.0]": 102.21385019455253, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python[TinyLlama-1.1B-Chat-v1.0]": 83.16836734693877, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_bounce[TinyLlama-1.1B-Chat-v1.0]": 105.64901086956522, - "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_fabric_memory[TinyLlama-1.1B-Chat-v1.0]": 104.76987434554974, - "disaggregated/test_disaggregated.py::test_disaggregated_perf_metrics[TinyLlama-1.1B-Chat-v1.0]": 91.85413349514563, - "disaggregated/test_disaggregated.py::test_disaggregated_python_transceiver_host_offload[TinyLlama-1.1B-Chat-v1.0]": 72.75782105263158, - "disaggregated/test_disaggregated.py::test_disaggregated_qwen3_32b_fp8[Qwen3/Qwen3-32B-FP8]": 151.65707692307694, - "disaggregated/test_disaggregated.py::test_disaggregated_single_gpu[TinyLlama-1.1B-Chat-v1.0]": 86.56471394230769, - "disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0]": 71.0492895522388, - "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer[False-TinyLlama-1.1B-Chat-v1.0]": 54.64942857142857, - "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer_missing_blocks[False-TinyLlama-1.1B-Chat-v1.0]": 53.03601818181818, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_cancel_gen_requests[TinyLlama-1.1B-Chat-v1.0]": 33.09288200589971, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_llama_context_capacity[False-False-DeepSeek-V3-Lite-fp8/fp8]": 157.21709316770185, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[False-TinyLlama-1.1B-Chat-v1.0]": 46.077697058823524, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[True-TinyLlama-1.1B-Chat-v1.0]": 38.63458439716312, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[False-TinyLlama-1.1B-Chat-v1.0]": 43.92117916666667, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[True-TinyLlama-1.1B-Chat-v1.0]": 40.47734232954546, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8]": 101.05138360175695, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-True-DeepSeek-V3-Lite-fp8/fp8]": 89.19943502824859, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-False-DeepSeek-V3-Lite-fp8/fp8]": 104.34513333333334, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-True-DeepSeek-V3-Lite-fp8/fp8]": 104.84779656160458, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-False-TinyLlama-1.1B-Chat-v1.0]": 40.79223037323037, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-True-TinyLlama-1.1B-Chat-v1.0]": 44.52552798053528, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-False-TinyLlama-1.1B-Chat-v1.0]": 50.73347846889952, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-True-TinyLlama-1.1B-Chat-v1.0]": 50.999568345323745, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-False-Qwen3-8B-FP8]": 66.79170639534884, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-True-Qwen3-8B-FP8]": 61.497651162790696, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-False-Qwen3-8B-FP8]": 81.84019393939394, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-True-Qwen3-8B-FP8]": 63.21660349854228, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[False-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 57.31423822714682, - "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[True-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 56.641159292035404, - "disaggregated/test_workers.py::test_workers_conditional_disaggregation_deepseek_v3_lite_bf16[DeepSeek-V3-Lite-bf16]": 140.14826804123712, - "disaggregated/test_workers.py::test_workers_kv_cache_aware_router[TinyLlama-1.1B-Chat-v1.0]": 133.58280288461538, - "disaggregated/test_workers.py::test_workers_kv_cache_events[TinyLlama-1.1B-Chat-v1.0]": 79.16608937198068, - "examples/test_ad_guided_decoding.py::test_autodeploy_guided_decoding_main_json": 108.65893142857144, - "examples/test_ad_speculative_decoding.py::test_autodeploy_eagle3_one_model_acceptance_rate[flashinfer-torch-simple]": 77.57904189944134, - "examples/test_ad_speculative_decoding.py::test_eagle_model_with_weights": 2.5963277777777773, - "examples/test_ad_speculative_decoding.py::test_nemotron_mtp_model_with_weights": 835.9901866295264, - "examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke": 989.6935238095238, - "examples/test_llm_api_with_mpi.py::test_llm_api_single_gpu_with_mpirun[TinyLlama-1.1B-Chat-v1.0]": 49.392453516969994, - "examples/test_ray.py::test_llm_inference_async_ray": 129.92340111420614, - "examples/test_ray.py::test_llm_inference_distributed_ray[tep2]": 160.54990721649486, - "examples/test_ray.py::test_llm_inference_distributed_ray[tp2]": 68.30646391752578, - "examples/visual_gen/test_visual_gen.py::test_cosmos3_example": 297.64562113402064, - "examples/visual_gen/test_visual_gen.py::test_flux1_example": 248.61007180851064, - "examples/visual_gen/test_visual_gen.py::test_flux1_lpips_against_golden": 53.898, - "examples/visual_gen/test_visual_gen.py::test_flux2_example": 295.3289088541667, - "examples/visual_gen/test_visual_gen.py::test_flux2_lpips_against_golden": 103.39257142857143, - "examples/visual_gen/test_visual_gen.py::test_ltx2_cuda_graph_lpips_matches_eager": 338.59385714285713, - "examples/visual_gen/test_visual_gen.py::test_ltx2_example": 379.62811936339523, - "examples/visual_gen/test_visual_gen.py::test_ltx2_lpips_against_golden": 65.055, - "examples/visual_gen/test_visual_gen.py::test_qwen_image_example": 231.70423376623376, - "examples/visual_gen/test_visual_gen.py::test_qwenimage_cuda_graph_lpips_against_golden": 68.01414285714286, - "examples/visual_gen/test_visual_gen.py::test_qwenimage_lpips_against_golden": 65.81428571428572, - "examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough": 271.3475490196078, - "examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart": 199.75915289982424, - "examples/visual_gen/test_visual_gen.py::test_wan21_t2v_lpips_against_golden": 21.5448, - "examples/visual_gen/test_visual_gen.py::test_wan22_t2v_lpips_against_golden": 90.2427, - "examples/visual_gen/test_visual_gen.py::test_wan_t2v_example": 726.5599819121447, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[attn2d_2x2]": 839.0274545454546, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[cfg2_ulysses2]": 261.68454545454546, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[ulysses4]": 607.2706666666667, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[cfg2_tp2]": 236.3876, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2]": 71.3225, - "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_ulysses2]": 302.0615, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_batch_generation": 20.855231805929918, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_cold_start": 20.51119398907104, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_field_completeness": 12.331842541436465, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_full_block_reuse": 10.812967032967034, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_long_context": 38.267905913978495, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_partial_block_reuse": 19.251983516483516, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_rapid_fire": 14.49196694214876, - "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_shared_prefix": 20.32607988980716, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens": 333.74045871559633, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens": 132.6506574074074, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_eviction": 37.61371028037383, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_batch_size_limited": 107.80071629213484, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_block_reuse": 121.58346723646723, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_eviction_block_reuse": 20.056470752089137, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_multi_request": 129.70799719887955, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_with_eviction": 23.063224657534246, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[cuda_graph]": 24.279370473537604, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[no_cuda_graph]": 20.66685714285714, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_overlap": 17.89529213483146, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[non_overlap]": 123.73225210084034, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[overlap]": 121.00675637393768, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_partial_block_reuse": 121.48944350282486, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_token_budget_limited": 124.12047701149424, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_v2_vs_v1_basic": 128.52438797814207, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_chunked_prefill": 194.13237709497207, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_eviction": 58.65875284090909, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_multi_adapter_v2": 131.8038, - "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_v2": 112.23774785100287, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[guaranteed-chunked]": 159.27757306590257, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[max-util-chunked]": 159.43676802507838, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[no-overlap-chunked]": 170.64738461538462, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-chunked]": 166.18679824561406, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-no-chunked]": 164.12044207317075, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[python-scheduler]": 165.64470674486805, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[swa-chunked]": 133.17204360465118, - "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix_smoke": 133.89418328445748, - "llmapi/test_llm_api_connector.py::test_connector_async_onboard[False]": 18.619161835748795, - "llmapi/test_llm_api_connector.py::test_connector_async_onboard[True]": 18.2084039408867, - "llmapi/test_llm_api_connector.py::test_connector_async_save[False]": 15.075786924939468, - "llmapi/test_llm_api_connector.py::test_connector_async_save[True]": 12.830610294117648, - "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[False]": 25.233555282555283, - "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[True]": 24.342429611650484, - "llmapi/test_llm_api_connector.py::test_connector_e2e_persistent_cache": 22.890816625916873, - "llmapi/test_llm_api_connector.py::test_connector_multi_request": 10.410875305623472, - "llmapi/test_llm_api_connector.py::test_connector_priorities": 12.587395061728396, - "llmapi/test_llm_api_connector.py::test_connector_priorities_default": 12.447050724637682, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[attention_dp]": 4.182096153846154, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[beam_search]": 10.050549636803876, - "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[host_offloading]": 11.549893975903615, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[False]": 14.680273170731708, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[True]": 14.689616504854367, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[False]": 11.245905569007263, - "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[True]": 11.27094362745098, - "llmapi/test_llm_api_connector.py::test_connector_simple[False]": 12.478055288461539, - "llmapi/test_llm_api_connector.py::test_connector_simple[True]": 12.50396805896806, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-bart-large-cnn]": 26.003712328767122, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-bart-large-cnn]": 25.386357954545456, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-bart-large-cnn]": 22.717285714285715, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 7.3287808641975305, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-bart-large-cnn]": 6.780027777777778, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-bart-large-cnn]": 14.974052459016393, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-bart-large-cnn]": 22.9215, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 22.552, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 6.945199084668192, - "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 8.355032, - "llmapi/test_llm_api_pytorch_bart.py::test_mbart_pytorch_generate_encoder_decoder_end_to_end": 22.4349012345679, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_mixtral_moe_routed_expert_fp8_multi_lora_varying_ranks[cudagraph]": 5400.0, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_mixtral_moe_routed_expert_fp8_multi_lora_varying_ranks[eager]": 5400.0, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[cudagraph]": 79.19040469208211, - "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[eager]": 91.1888, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-t5-small]": 6.15115313225058, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 11.990981481481482, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-xl]": 38.03971428571428, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-t5-small]": 21.127428571428574, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-base]": 22.332, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-small]": 6.14457911908646, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-t5-small]": 12.96089082278481, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-t5-small]": 5.369179566563467, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-tp2-t5-small]": 17.288162393162395, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-off-greedy-byt5-small]": 13.017768595041321, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 21.729333333333333, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-t5-small]": 20.64085714285714, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-t5-small]": 5.132682098765432, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 21.11866666666667, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-t5-small]": 20.56142857142857, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 21.302833333333332, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-t5-small]": 20.521285714285714, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 20.4565, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-t5-small]": 19.902714285714286, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 20.3345, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-t5-small]": 19.869142857142858, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_context_generation_batch": 5.753551162790698, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-flan-t5-small]": 5.0157716535433075, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-t5-small]": 4.783761061946902, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-t5-small]": 5.329487640449438, - "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-t5-small]": 7.091325396825397, - "llmapi/test_llm_examples.py::test_llmapi_chat_example": 70.07263636363636, - "llmapi/test_llm_examples.py::test_llmapi_example_distributed_tp2": 56.021, - "llmapi/test_llm_examples.py::test_llmapi_example_guided_decoding": 65.55775636811832, - "llmapi/test_llm_examples.py::test_llmapi_example_inference": 71.85408210180624, - "llmapi/test_llm_examples.py::test_llmapi_example_inference_async": 53.34649383730485, - "llmapi/test_llm_examples.py::test_llmapi_example_inference_async_streaming": 48.074224548440064, - "llmapi/test_llm_examples.py::test_llmapi_example_logits_processor": 47.359884646962236, - "llmapi/test_llm_examples.py::test_llmapi_example_multilora": 61.29332004930156, - "llmapi/test_llm_examples.py::test_llmapi_kv_cache_connector[Qwen2-0.5B]": 91.71036363636364, - "llmapi/test_llm_examples.py::test_llmapi_quickstart": 48.46666379310345, - "llmapi/test_llm_examples.py::test_llmapi_quickstart_atexit": 319.7374347826087, - "llmapi/test_llm_examples.py::test_llmapi_server_example": 157.16363636363636, - "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_mtp": 244.0717337962963, - "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_ngram": 95.34129156908665, - "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_context": 1.0290494505494505, - "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_generation": 0.967483606557377, - "perf/host_perf/test_module_sampler.py::test_sampler_update_greedy[greedy_bs8]": 1.2933342618384402, - "perf/host_perf/test_module_sampler.py::test_sampler_update_stop_words[stopwords_bs32]": 1.4042594594594595, - "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_gen_only_bs8]": 0.925497191011236, - "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_mixed_32gen_4ctx]": 1.055345945945946, - "perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:8192]": 255.99544495412843, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 2016.2651538461537, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 379.92807692307696, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 1608.9949, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 951.954875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 371.3276923076923, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2365.190263414634, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 434.1525384615384, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 2853.0363846153846, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 376.83946153846153, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 2775.1226923076924, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 203.80992307692307, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 576.1800634146341, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1397.571923076923, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2195.1416153846153, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 501.6934615384616, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2228.081307317073, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 300.3468461538461, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 416.32678571428573, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 4561.989421052632, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2140.55235, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1500.5563333333332, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 2060.203, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1038.6396666666667, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 2913.0081578947365, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 423.3564375, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 1640.4845625, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1264.9850526315788, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1965.8842105263157, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 470.33705263157896, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 1331.24528125, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_tep8_mtp3]": 565.2403448275862, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 664.1381538461538, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 678.5075, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_dep4_mtp1_8k1k]": 916.9552524752474, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_1k8k]": 957.3329230769231, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_8k1k]": 573.0860742574257, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_1k8k]": 605.0051538461539, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 416.47730917874395, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_ad_blackwell-r1_fp8_ad_ws8_1k1k]": 755.3983636363637, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_blackwell-r1_fp8_tp8_6k1k]": 1130.9022307692308, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_dep8_mtp1_8k1k]": 817.33225, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_tep8_mtp3_8k1k]": 492.7644166666667, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_dep4_mtp1_8k1k]": 984.8252307692308, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_tep4_mtp3_8k1k]": 405.94523076923076, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_deepseek_v32_fp4_2_nodes_grace_blackwell-dsv32_fp4_dep8_trtllm_lpc_mnnvl_8k1k]": 498.78161538461535, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_adp_2k1k]": 574.2704166666666, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_235b_a22b_fp8_hopper-qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1k]": 957.4866363636363, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_32b_fp8_hopper-qwen3_32b_fp8_tp2_6k1k]": 776.424090909091, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 630.5806153846154, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_tep8_mtp3_8k1k]": 581.6817692307693, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 590.7949166666666, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 606.2082307692308, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_dep4_1k8k]": 1336.2756923076922, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tep2_1k8k]": 811.5083846153847, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 540.0438461538462, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp2_1k8k]": 395.9511538461539, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_deepseek_v3_lite-v3lite_fp8_bs8_128_256]": 573.4216694677872, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b-llama8b_fp16_bs8_128_256]": 268.8767824858757, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b_spec_decode-llama8b_spec_bs1_128_128]": 150.8482967032967, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_2_nodes_grace_blackwell-k25_thinking_fp4_dep8_32k8k]": 1488.879923076923, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_32k8k]": 1444.9784166666668, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_8k1k]": 766.74, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_32k8k]": 657.536, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_8k1k]": 445.1063846153846, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_dep4_8k1k]": 841.3804830917875, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_tep4_8k1k]": 457.48476923076925, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_1000_1000]": 463.101, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_512_32]": 252.52316666666667, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]": 684.3395384615385, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]": 580.3056153846154, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep4_mtp3_8k1k]": 386.42491666666666, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tp4_8k1k]": 461.85966666666667, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_ad_blackwell-super_ad_ws4_1k1k]": 564.4200416666666, - "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_mtp_ad_blackwell-super_mtp_ad_ws4_1k1k]": 532.4755454545455, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 4133.825, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2472.7274545454547, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 3231.6324285714286, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 3142.5304615384616, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 851.2984308510638, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2151.5645, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2125.6215384615384, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 8068.51475, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2211.1548095238095, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1339.5592666666666, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 2621.412888888889, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1377.9775555555555, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1276.3097857142857, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 3251.7198439716312, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2548.912, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2466.507727272727, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 842.7718000000001, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 889.5297857142857, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 517.3953529411765, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 544.902, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb256_mtp3_ccb-NIXL]": 883.5191538461539, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 312.7136666666667, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 264.12928571428574, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 828.4586666666667, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 488.6693076923077, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 3147.871157894737, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 836.2682272727272, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 1035.4665454545454, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 607.4167, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 586.0575, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1462.754, - "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 457.9464835680751, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-flux2_blackwell-flux2_fp8_cfg1_ulysses4_teacache_on]": 360.2670909090909, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_i2v_cfg2_ulysses4_compile_on]": 536.7069, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_t2v_cfg2_ulysses4_compile_on]": 474.23836363636366, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_nvfp4_i2v_cfg2_ulysses4_compile_on]": 458.0541818181818, - "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan21_t2v_14b_blackwell-wan21_14b_nvfp4_trtllm_cfg2_ulysses4_teacache_on]": 513.6001818181818, - "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp1_4instances]": 106.18771875, - "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp2_2instances]": 109.6773125, - "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-GUARANTEED_NO_EVICT-pytorch-stress-test]": 737.0806363636364, - "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-MAX_UTILIZATION-pytorch-stress-test]": 664.5234545454546, - "test_e2e.py::test_get_ci_container_port": 1.3634927884615387, - "test_e2e.py::test_openai_chat_example[pytorch]": 575.1747253012048, - "test_e2e.py::test_openai_chat_guided_decoding[meta-llama/Llama-3.1-8B-Instruct]": 74.3235283018868, - "test_e2e.py::test_openai_chat_guided_decoding[openai/gpt-oss-120b]": 183.251, - "test_e2e.py::test_openai_chat_harmony": 174.82321751412428, - "test_e2e.py::test_openai_chat_harmony_perf_metrics": 160.05857558139536, - "test_e2e.py::test_openai_chat_multimodal_example": 130.39714077669902, - "test_e2e.py::test_openai_chat_with_logit_bias[torch_sampler]": 86.90059294117647, - "test_e2e.py::test_openai_chat_with_logit_bias[trtllm_sampler]": 87.14738461538461, - "test_e2e.py::test_openai_completions_example[pytorch]": 261.67624220623503, - "test_e2e.py::test_openai_completions_with_logit_bias[torch_sampler]": 84.34349065420561, - "test_e2e.py::test_openai_completions_with_logit_bias[trtllm_sampler]": 84.66151643192488, - "test_e2e.py::test_openai_health": 72.27581818181818, - "test_e2e.py::test_openai_kv_cache_contamination": 1411.96477, - "test_e2e.py::test_openai_lora": 185.54133078880406, - "test_e2e.py::test_openai_misc_example[pytorch]": 350.13566343825664, - "test_e2e.py::test_openai_mmencoder_example": 136.0997242206235, - "test_e2e.py::test_openai_perf_metrics": 86.54126086956522, - "test_e2e.py::test_openai_post_processor": 372.86061686746984, - "test_e2e.py::test_openai_prometheus": 123.17233812949641, - "test_e2e.py::test_openai_reasoning[pytorch]": 257.26778132678135, - "test_e2e.py::test_openai_responses": 311.0094705882353, - "test_e2e.py::test_openai_responses_entrypoint": 107.0407848410758, - "test_e2e.py::test_openai_tool_call": 89.66446634615384, - "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-120B-gpt_oss/gpt-oss-120b]": 1067.7551818181819, - "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-20B-gpt_oss/gpt-oss-20b]": 214.3322994923858, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B]": 219.58391370558374, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]": 127.58688772845953, - "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta-Llama-3.1-8B]": 153.47955013192612, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]": 466.13843444730077, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_fp8_hf-Qwen3/saved_models_Qwen3-30B-A3B_fp8_hf]": 256.2311825192802, - "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_nvfp4_hf-Qwen3/saved_models_Qwen3-30B-A3B_nvfp4_hf]": 178.14921391752577, - "test_e2e.py::test_ptp_quickstart_advanced_bs1": 96.76857000000001, - "test_e2e.py::test_ptp_quickstart_advanced_deepseek_v3_lite_4gpus_adp_balance[DeepSeek-V3-Lite-FP8-DeepSeek-V3-Lite/fp8]": 94.38419387755103, - "test_e2e.py::test_ptp_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 80.07500583090379, - "test_e2e.py::test_ptp_quickstart_advanced_mixed_precision": 90.66226975476839, - "test_e2e.py::test_ptp_quickstart_advanced_mtp[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 190.38033875338755, - "test_e2e.py::test_ptp_quickstart_advanced_mtp_eagle[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 280.69667039106145, - "test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct]": 70.9769423631124, - "test_e2e.py::test_ptp_quickstart_bert[TRTLLM-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 23.737279936558288, - "test_e2e.py::test_ptp_quickstart_bert[VANILLA-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 7.569804761904762, - "test_e2e.py::test_trtllm_bench_help_sanity[meta-llama/Llama-3.1-8B]": 48.68448973607038, - "test_e2e.py::test_trtllm_bench_invalid_token_pytorch[TinyLlama-1.1B-Chat-v1.0-TinyLlama-1.1B-Chat-v1.0]": 79.17544525547444, - "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-non-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.94045479452055, - "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.49601497005989, - "test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b]": 120.57213636363636, - "test_e2e.py::test_trtllm_bench_mgmn": 110.8202, - "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-False-False]": 114.68200189393939, - "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-instruct-hf-fp8-True-True]": 69.32486930091186, - "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-]": 122.67552884615384, - "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-enable_request_rate]": 37.874408823529414, - "test_e2e.py::test_trtllm_serve_example": 178.86436363636363, - "test_e2e.py::test_trtllm_serve_lora_example": 94.39688040712468, - "test_e2e.py::test_trtllm_serve_multimodal_example": 178.93932289156626, - "test_e2e.py::test_trtllm_serve_top_logprobs[pytorch]": 86.90236144578313, - "test_fmha.py::test_fmha": 2348.44875, - "thirdparty/test_cmake_third_party.py::test_cmake_listfiles": 1.3952367149758453, - "thirdparty/test_git_modules.py::test_gitmodules": 1.2932296650717703, - "unittest/_torch/attention": 1499.0411369346734, - "unittest/_torch/attention --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1461.3758554216868, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_kernel.py": 26.301077142857142, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_module.py": 116.55956629834255, - "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_tf32.py": 21.41217094017094, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_cache_manager.py": 52.60678431372549, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_indices_transform.py": 50.76427808988764, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_o_proj.py": 37.86423978201635, - "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_sparse_mla.py": 52.061539944903586, - "unittest/_torch/attention/sparse/dsa/test_dsa_fp4_indexer.py": 26.329861386138614, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_cute_dsl_fp8_indexer_q_gemm_rope_fp4_matches_unfused": 45.63755555555556, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_dsv32_prefill_shape": 19.948702970297028, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_matches_deepgemm": 19.606653465346536, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_noncontiguous_split": 2.04908071278826, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_shape_dispatch": 19.43559405940594, - "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_indexer_k_cache_gather_contiguous_fp4": 19.109356435643562, - "unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py": 701.1912252747252, - "unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py": 348.6813176470588, - "unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1584.264129213483, - "unittest/_torch/attention/sparse/test_sparse_mla_forward.py": 111.27268644067796, - "unittest/_torch/attention/test_attention_mla.py": 135.2430928319624, - "unittest/_torch/auto_deploy": 26.348274924471298, - "unittest/_torch/compilation": 26.10724778761062, - "unittest/_torch/custom_ops/test_deepseek_v4_q_norm.py": 22.96618384401114, - "unittest/_torch/debugger": 20.013879710144927, - "unittest/_torch/distributed": 20.335233009708737, - "unittest/_torch/executor": 337.5190529327611, - "unittest/_torch/executor/test_async_transfer_manager.py": 27.127968215158926, - "unittest/_torch/executor/test_disagg_index_mapper_early_release.py": 27.426251781472686, - "unittest/_torch/executor/test_error_classification.py": 25.870426829268293, - "unittest/_torch/executor/test_hang_detector_kill.py": 42.569074018877295, - "unittest/_torch/executor/test_kv_cache_budget_split.py": 27.206930952380954, - "unittest/_torch/executor/test_kv_cache_compression_manager.py": 27.282539024390243, - "unittest/_torch/executor/test_kv_cache_estimation.py": 27.37214423076923, - "unittest/_torch/executor/test_kv_cache_v2_capacity_only.py": 25.55210606060606, - "unittest/_torch/executor/test_kv_pool_rebalance.py": 27.236201456310678, - "unittest/_torch/executor/test_model_loader_mx.py": 23.38812343201204, - "unittest/_torch/executor/test_overlap_scheduler.py": 796.4592737430168, - "unittest/_torch/executor/test_scheduler_serializable_output.py": 25.658269417475726, - "unittest/_torch/flashinfer/test_trtllm_flashinfer_symbol_collision.py": 1003.646782234957, - "unittest/_torch/lora": 21.104300771208226, - "unittest/_torch/memory": 20.41897507788162, - "unittest/_torch/misc": 217.8719063444109, - "unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy": 35.712514851485146, - "unittest/_torch/modeling -k \"modeling_gemma3\"": 250.2099910714286, - "unittest/_torch/modeling -k \"modeling_gpt_oss\"": 99.71302683615819, - "unittest/_torch/modeling -k \"modeling_llama\"": 124.79015498652292, - "unittest/_torch/modeling -k \"modeling_mixtral\"": 70.61155039787799, - "unittest/_torch/modeling -k \"modeling_nemotron_nano_v2_vl\"": 426.0704334140436, - "unittest/_torch/modeling -k \"modeling_nemotron_nas\"": 38.65103496503497, - "unittest/_torch/modeling -k \"modeling_out_of_tree\"": 146.20377011494253, - "unittest/_torch/modeling -k \"modeling_phi3\"": 33.42191375291375, - "unittest/_torch/modeling -k \"modeling_qwen\"": 707.234, - "unittest/_torch/modeling -k \"modeling_qwen_moe\"": 292.4790489510489, - "unittest/_torch/modeling -k \"modeling_siglip\"": 130.33033101851854, - "unittest/_torch/modeling -k \"modeling_speculative\"": 27.073209790209788, - "unittest/_torch/modeling -k \"modeling_vila\"": 42.617395784543326, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_multimodal_26b_dummy": 188.08098222222222, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_26b_dummy": 160.7302222222222, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_31b_dummy": 161.78340322580647, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy": 148.25198002853068, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[256]": 173.58835267857143, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[512]": 192.90066968325795, - "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e4b_dummy": 195.04240348330913, - "unittest/_torch/modeling/test_gemma4_multimodal.py": 24.872615803814714, - "unittest/_torch/modeling/test_modeling_afmoe.py": 65.08802259887005, - "unittest/_torch/modeling/test_modeling_cohere2.py": 60.729254196642685, - "unittest/_torch/modeling/test_modeling_cosmos3.py::TestCosmos3::test_all": 25.520618604651162, - "unittest/_torch/modeling/test_modeling_deepseekv4.py": 57.04133485193622, - "unittest/_torch/modeling/test_modeling_exaone_moe.py": 255.23339285714286, - "unittest/_torch/modeling/test_modeling_gemma4.py": 77.90781034482758, - "unittest/_torch/modeling/test_modeling_gemma4_unified.py": 20.19156862745098, - "unittest/_torch/modeling/test_modeling_llava_next.py::TestLlavaNext::test_all": 29.02741046511628, - "unittest/_torch/modeling/test_modeling_llava_next.py::test_llava_next_expand_prompt_token_ids_for_mm": 14.212421658986175, - "unittest/_torch/modeling/test_modeling_mistral.py": 37.79800717703349, - "unittest/_torch/modeling/test_modeling_nemotron_h.py::test_nemotron_h_sanity": 156.58529545454545, - "unittest/_torch/modeling/test_modeling_parakeet.py": 44.25659903381643, - "unittest/_torch/modeling/test_modeling_pixtral.py": 35.958952267303104, - "unittest/_torch/modeling/test_modeling_pixtral.py::test_tensor_parallelism": 43.275, - "unittest/_torch/modeling/test_modeling_qwen2_5vl.py::TestQwen2_5_VL::test_all": 76.14861520467836, - "unittest/_torch/modeling/test_modeling_qwen3_5_vl.py::TestQwen3_5VL::test_all": 73.7980761904762, - "unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py::TestQwen3_5MoeVL::test_all": 80.15000454545455, - "unittest/_torch/modeling/test_modeling_qwen3vl.py::TestQwen3VL::test_all": 96.30822105263158, - "unittest/_torch/modeling/test_modeling_qwen3vl.py::test_qwen3vl_init_preserves_caller_quant_config": 16.023488966318236, - "unittest/_torch/modeling/test_modeling_qwen3vl_moe.py::TestQwen3VLMoe::test_all": 82.7298538011696, - "unittest/_torch/modeling/test_modeling_radio.py": 29.436779904306217, - "unittest/_torch/modeling/test_modeling_step3p7.py": 28.5279375, - "unittest/_torch/modeling/test_modeling_step3p7vl.py": 38.826083720930235, - "unittest/_torch/modeling/test_multimodal_encoder_graph.py": 20.719585294117646, - "unittest/_torch/modeling/test_multimodal_encoder_mixin.py": 25.568744578313254, - "unittest/_torch/modeling/test_nemotron_nano_preprocessing.py": 62.31733732057416, - "unittest/_torch/models/checkpoints": 62.52584578313253, - "unittest/_torch/modules/dwdp/test_dwdp_fixup_moe_backends.py": 26.235854066985645, - "unittest/_torch/modules/dwdp/test_dwdp_manager.py": 25.706846715328467, - "unittest/_torch/modules/dwdp/test_dwdp_mapping.py": 25.72155339805825, - "unittest/_torch/modules/dwdp/test_dwdp_peer_ranges.py": 25.59992048192771, - "unittest/_torch/modules/fused_moe/test_deepgemm_fused_expand_quant.py": 22.875341658341657, - "unittest/_torch/modules/fused_moe/test_deepgemm_fused_gather_finalize.py": 22.920138532110094, - "unittest/_torch/modules/fused_ops/test_fused_qk_norm_rope_gate.py": 23.7237, - "unittest/_torch/modules/fused_ops/test_gelu_tanh_mul_fp4_quant.py": 24.07451566951567, - "unittest/_torch/modules/fused_ops/test_rmsnorm_fp4_quant.py": 23.92312912087912, - "unittest/_torch/modules/fused_ops/test_rmsnorm_residual_add.py": 24.953952802359883, - "unittest/_torch/modules/mamba": 740.4738101265823, - "unittest/_torch/modules/moe/test_cute_dsl_b12x_moe_backend.py": 21.06453807106599, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 91.753, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL\"": 342.06576923076926, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTLASS\"": 1208.6856129032258, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 46.524699999999996, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM\"": 62.843666666666664, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DENSEGEMM\"": 644.2885555555555, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_CUTEDSL\"": 243.494, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_DEEPGEMM\"": 27.79185714285714, - "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"TRTLLM\"": 123.4027619047619, - "unittest/_torch/modules/moe/test_moe_backend.py::test_trtllm_bf16_unquantized_moe": 899.1003994252874, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm": 206.84880319148937, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_boundary": 196.21933333333334, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_postquant": 118.7101182795699, - "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_nccl_ep_cuda_graph_replay_uses_updated_routing": 41.391349999999996, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTEDSL\"": 106.04984158415841, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES and not W4A8 and not MXFP8\"": 82.33142857142856, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES\"": 71.5973, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8_BLOCK_SCALES\"": 60.02572727272727, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and MXFP8 and not W4A8\"": 70.1191836734694, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and NVFP4\"": 86.05340816326532, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A16_MXFP4\"": 81.69265979381443, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_AWQ\"": 68.79285, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_FP8\"": 81.08657575757576, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_MXFP8\"": 150.906099009901, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W8A16\"": 91.86447959183673, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 84.99442574257425, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MARLIN\"": 81.10676923076923, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_CUTEDSL\"": 59.32976288659793, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_DEEPGEMM\"": 51.67359595959596, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and FP8_BLOCK_SCALES\"": 63.96949504950495, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and NVFP4 and not W4A8\"": 59.18147524752475, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A16_MXFP4\"": 87.15429591836735, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_MXFP4_MXFP8\"": 51.86418811881188, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_NVFP4_FP8\"": 157.12869306930693, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 64.47758, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 47.0703, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 54.87052, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 80.27069, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.16604, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_CUTEDSL-quant=NVFP4-routing=DeepSeekV3]": 55.19516326530612, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_DEEPGEMM-quant=W4A8_MXFP4_MXFP8-routing=DeepSeekV3]": 56.76075510204081, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 26.64341, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 26.58097, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 27.330119999999997, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 48.381660000000004, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.22708, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.49764, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb": 48.43913725490196, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_CUTEDSL\"": 56.218061224489794, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_DEEPGEMM\"": 52.125959183673466, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"not None\"": 25.63853608247423, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 46.70682, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 47.37518, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 47.52538, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 91.13814722222222, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and None\"": 27.01264705882353, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and not None\"": 151.68869058295965, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 49.13182647058823, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DENSEGEMM\"": 463.5970454545455, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_CUTEDSL\"": 175.77482905982907, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_DEEPGEMM\"": 42.02627173913043, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 65.25673630136987, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 32.85159831460675, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 30.820144329896905, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 29.589710743801653, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 30.251238754325257, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.18373224043716, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.542106529209626, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 33.902055401662054, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 29.093491408934707, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 57.66461690140845, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.324429906542058, - "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.float16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.324984813084114, - "unittest/_torch/modules/test_awq_quantization.py": 20.191866873065017, - "unittest/_torch/modules/test_engram.py": 102.07443352601156, - "unittest/_torch/modules/test_fp4_num_tokens_slice.py": 21.468339943342777, - "unittest/_torch/modules/test_fused_activation_quant.py": 21.324547932330827, - "unittest/_torch/modules/test_fused_add_rms_norm_quant.py": 21.579973869346734, - "unittest/_torch/modules/test_fused_rmsnorm_fp4_quantize.py": 28.061774853801168, - "unittest/_torch/modules/test_gemma4_fused_qkv_prep.py": 25.013674285714284, - "unittest/_torch/modules/test_group_rmn_norm.py": 26.816904549509367, - "unittest/_torch/modules/test_mhc.py": 29.68183914728682, - "unittest/_torch/modules/test_mla_helix.py": 52.47190936106984, - "unittest/_torch/modules/test_moe_host_sharer.py": 22.169184415584414, - "unittest/_torch/modules/test_moe_load_balancer.py": 22.058864864864862, - "unittest/_torch/modules/test_moe_routing.py": 47.47694806421152, - "unittest/_torch/modules/test_rotary_embedding.py": 22.013752016129033, - "unittest/_torch/modules/test_triton_linear.py": 21.429786885245903, - "unittest/_torch/modules/tests_lora_modules": 334.8057459807074, - "unittest/_torch/multi_gpu -m \"not post_merge\"": 340.56549333333334, - "unittest/_torch/multi_gpu_modeling -k \"deepseek\"": 311.1584347826087, - "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp1-bf16-trtllm-deepseekv3_lite]": 87.91341395348837, - "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp4-bf16-trtllm-deepseekv3_lite]": 115.07457209302325, - "unittest/_torch/multimodal -k \"not nemotron_nano_v2_vl_fp8\"": 914.8516534653465, - "unittest/_torch/multimodal/test_mm_encoder_standalone.py -k \"qwen3_30b_a3b_fp8\"": 386.13995362318843, - "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu2\"": 229.27255670103094, - "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu4\"": 470.16044791666667, - "unittest/_torch/ray_orchestrator/multi_gpu/test_inflight_weight_update.py": 83.6284947368421, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part0\"": 606.4082, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part1\"": 434.8369, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part2\"": 604.1570294117647, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part3\"": 521.5557475728156, - "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part4\"": 1015.58, - "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp1_4instances]": 278.0227083333333, - "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp2_2instances]": 293.1756666666667, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part0\"": 286.79657731958764, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part1\"": 287.2486494845361, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part2\"": 155.7070206185567, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part3\"": 156.41694845360826, - "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part4\"": 214.8349175257732, - "unittest/_torch/ray_orchestrator/single_gpu/test_cache_transceiver_comm.py": 50.02742076502732, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_sleep.py": 116.42223463687151, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part0\"": 718.5405393258427, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part1\"": 676.5004831460674, - "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part2\"": 496.98034550561795, - "unittest/_torch/sampler -k \"not test_speculative_d2h_parity_real_predictor\"": 629.0897239263804, - "unittest/_torch/sampler/test_beam_search.py": 466.1707026378897, - "unittest/_torch/sampler/test_beam_search_speculative_d2h.py": 130.66764, - "unittest/_torch/sampler/test_logits_logprobs.py": 274.3431100478469, - "unittest/_torch/sampler/test_torch_sampler.py": 207.59631818181816, - "unittest/_torch/sampler/test_trtllm_sampler.py": 98.30468292682926, - "unittest/_torch/speculative/hw_agnostic": 841.041305882353, - "unittest/_torch/speculative/test_eagle3.py": 657.948134218289, - "unittest/_torch/speculative/test_rejection_buffers_guard.py": 20.309, - "unittest/_torch/test_model_config.py": 25.657892523364488, - "unittest/_torch/test_torch_multi_arange.py": 46.24331719128329, - "unittest/_torch/thop/parallel": 1885.2997194670613, - "unittest/_torch/thop/parallel/test_finegrained_mixed_dtype_gemm.py": 28.596906801007556, - "unittest/_torch/thop/parallel/test_w4a16_linear.py": 21.82673299748111, - "unittest/_torch/thop/parallel/test_w4a8_linear.py": 20.717201511335013, - "unittest/_torch/thop/parallel/test_weight_only_quant_gemm.py": 58.659168765743075, - "unittest/_torch/thop/parallel/test_weight_only_quant_linear.py": 26.54437279596977, - "unittest/_torch/thop/parallel_hw_agnostic": 470.2435887573964, - "unittest/_torch/thop/serial": 52.232433867735466, - "unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py": 305.10693913043474, - "unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py": 218.1122830188679, - "unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py": 231.8527663551402, - "unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py": 192.40505319148937, - "unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py": 146.66659047619046, - "unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py": 70.97888679245283, - "unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py": 121.93633644859813, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py": 44.38675531914893, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py": 84.08998095238096, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py": 64.73582075471698, - "unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py": 250.16000934579438, - "unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py": 206.29218279569892, - "unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py": 296.3244380952381, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py": 66.38530188679246, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py": 392.43442990654205, - "unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py": 134.25890425531915, - "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py": 182.28520175438595, - "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_multinode.py": 22.10380172413793, - "unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py": 75.1850186915888, - "unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py": 308.29408510638297, - "unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py": 180.6913904761905, - "unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py": 455.3819038461539, - "unittest/_torch/visual_gen/test_attention_cute_dsl.py": 136.91421487603304, - "unittest/_torch/visual_gen/test_attention_cute_dsl_vsa.py": 43.35358474576271, - "unittest/_torch/visual_gen/test_attention_integration.py": 97.10202133333334, - "unittest/_torch/visual_gen/test_attention_perf.py": 210.67039617486338, - "unittest/_torch/visual_gen/test_attention_trtllm_sage.py": 189.85255145118734, - "unittest/_torch/visual_gen/test_cache_dit.py": 147.32329178470255, - "unittest/_torch/visual_gen/test_cosmos3_pipeline.py": 161.16777876106195, - "unittest/_torch/visual_gen/test_cosmos3_transformer.py": 98.51558529411764, - "unittest/_torch/visual_gen/test_flux_attention.py": 26.92808579088472, - "unittest/_torch/visual_gen/test_flux_infer.py": 25.655417233560094, - "unittest/_torch/visual_gen/test_flux_pipeline.py": 531.3778176795581, - "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxCombinedOptimizations::test_all_optimizations_combined": 77.22944036697247, - "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxParallelism::test_ulysses_2gpu_correctness": 64.36324778761062, - "unittest/_torch/visual_gen/test_flux_transformer.py": 38.82068067226891, - "unittest/_torch/visual_gen/test_ltx2_attention.py": 24.79130769230769, - "unittest/_torch/visual_gen/test_ltx2_pipeline.py": 717.9058804347826, - "unittest/_torch/visual_gen/test_ltx2_transformer.py": 27.601359173126614, - "unittest/_torch/visual_gen/test_model_loader.py": 46.429017441860466, - "unittest/_torch/visual_gen/test_quant_ops.py": 21.580876832844574, - "unittest/_torch/visual_gen/test_qwen_image_infer.py": 25.610972413793103, - "unittest/_torch/visual_gen/test_qwen_image_pipeline.py": 25.810387096774193, - "unittest/_torch/visual_gen/test_teacache.py": 20.69974858757062, - "unittest/_torch/visual_gen/test_tensor_payload.py": 25.807934240362812, - "unittest/_torch/visual_gen/test_trtllm_serve_e2e.py": 493.1424769647696, - "unittest/_torch/visual_gen/test_trtllm_serve_endpoints.py": 29.099208108108108, - "unittest/_torch/visual_gen/test_visual_gen_args.py": 21.59079120879121, - "unittest/_torch/visual_gen/test_visual_gen_params.py": 26.508512376237626, - "unittest/_torch/visual_gen/test_visual_gen_utils.py": 21.791359890109888, - "unittest/_torch/visual_gen/test_wan21_i2v_pipeline.py": 223.45977393617022, - "unittest/_torch/visual_gen/test_wan21_i2v_teacache.py": 348.9664619883041, - "unittest/_torch/visual_gen/test_wan21_t2v_pipeline.py": 192.33468047337277, - "unittest/_torch/visual_gen/test_wan21_t2v_teacache.py": 250.91146764705883, - "unittest/_torch/visual_gen/test_wan21_t2v_teacache_user_coefficients.py": 55.60930623306233, - "unittest/_torch/visual_gen/test_wan22_i2v_pipeline.py": 305.1618966480447, - "unittest/_torch/visual_gen/test_wan22_i2v_teacache.py": 149.2714253521127, - "unittest/_torch/visual_gen/test_wan22_t2v_pipeline.py": 287.8022845303867, - "unittest/_torch/visual_gen/test_wan22_t2v_teacache.py": 138.68507407407407, - "unittest/_torch/visual_gen/test_wan22_ti2v_5b_pipeline.py": 132.64411436170212, - "unittest/_torch/visual_gen/test_wan_transformer.py": 86.32018313953489, - "unittest/_torch/visual_gen/test_wan_vae.py": 66.85342215568862, - "unittest/_torch/visual_gen/test_wan_vsa_pipeline.py": 132.44576832844575, - "unittest/_torch/visual_gen/test_warmup.py": 21.973272237196763, - "unittest/_torch/weight_sharing": 25.870444964871197, - "unittest/api_stability": 29.708095823095825, - "unittest/auto_deploy/multigpu/compile": 43.853566666666666, - "unittest/auto_deploy/multigpu/smoke": 357.5653522012579, - "unittest/auto_deploy/multigpu/transformations": 2098.098430107527, - "unittest/auto_deploy/multigpu/transformations/library/test_allreduce_residual_rmsnorm_fusion.py -k \"strategy_auto\"": 119.75213402061856, - "unittest/auto_deploy/multigpu/transformations/library/test_tp_sharding.py::test_moe_tp_shard_nvfp4": 69.07451546391754, - "unittest/auto_deploy/singlegpu/compile": 68.55923552123552, - "unittest/auto_deploy/singlegpu/custom_ops": 522.1245855855856, - "unittest/auto_deploy/singlegpu/custom_ops/attention/test_triton_attention.py::TestSDPADispatch": 38.268032258064515, - "unittest/auto_deploy/singlegpu/custom_ops/mamba/test_flashinfer_mamba_cached_op.py": 47.28830913978495, - "unittest/auto_deploy/singlegpu/custom_ops/moe/test_ad_moe_op.py": 29.06793530997305, - "unittest/auto_deploy/singlegpu/custom_ops/moe/test_trtllm_moe.py": 40.64752291105122, - "unittest/auto_deploy/singlegpu/custom_ops/quantization/test_quant.py": 50.19106738544475, - "unittest/auto_deploy/singlegpu/mlir": 29.818291828793775, - "unittest/auto_deploy/singlegpu/models": 228.87805836575873, - "unittest/auto_deploy/singlegpu/shim": 91.26811240310077, - "unittest/auto_deploy/singlegpu/smoke": 1713.564464993395, - "unittest/auto_deploy/singlegpu/smoke/test_ad_build_small_single.py -k \"Nemotron-3-Nano-30B-A3B-FP8 or Nemotron-Nano-3-30B-A3.5B-dev or Llama-4-Scout\"": 249.2459164420485, - "unittest/auto_deploy/singlegpu/smoke/test_ad_speculative_decoding.py": 210.74064480874318, - "unittest/auto_deploy/singlegpu/transformations": 320.5240232858991, - "unittest/auto_deploy/singlegpu/transformations/library/test_fuse_relu2_quant_nvfp4.py": 36.78159029649596, - "unittest/auto_deploy/singlegpu/transformations/library/test_moe_fusion.py": 66.3970269541779, - "unittest/auto_deploy/singlegpu/transformations/library/test_nvfp4_swiglu.py": 34.473213513513514, - "unittest/auto_deploy/singlegpu/transformations/library/test_quant_fusion.py": 39.26718699186991, - "unittest/auto_deploy/singlegpu/transformations/library/test_quant_moe.py": 31.283571815718158, - "unittest/auto_deploy/singlegpu/transformations/library/test_quantization.py": 33.57898641304348, - "unittest/auto_deploy/singlegpu/transformations/library/test_trtllm_gen_diag.py": 41.16601630434782, - "unittest/auto_deploy/singlegpu/utils": 45.979049222797926, - "unittest/auto_deploy/standalone": 719.8760111731843, - "unittest/bindings": 41.77800726392252, - "unittest/disaggregated/region/test_aux.py": 20.04190855457227, - "unittest/disaggregated/region/test_block.py": 23.17737385620915, - "unittest/disaggregated/region/test_page.py": 20.239856756756755, - "unittest/disaggregated/region/test_region.py": 20.012345047923322, - "unittest/disaggregated/test_agent_multi_backends.py": 82.98323180592992, - "unittest/disaggregated/test_bounce.py": 23.589685828877005, - "unittest/disaggregated/test_cache_reuse_adapter.py": 25.585138888888892, - "unittest/disaggregated/test_cache_transceiver_harness.py": 56.23020353982301, - "unittest/disaggregated/test_cache_transceiver_harness_report.py": 21.828332378223493, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and no_window\"": 189.49816756756758, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and uniform_window\"": 171.13203492063494, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and no_window\"": 228.43785139318885, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and uniform_window\"": 243.70949554896143, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and vswa\"": 221.75007692307693, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v1\"": 53.39233522727273, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v2\"": 64.50010810810811, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"no_window\"": 163.61731012658228, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"uniform_window\"": 150.50420743034056, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"v2 and vswa\"": 102.53459940652819, - "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_v1_dsa_indexer": 49.56948913043478, - "unittest/disaggregated/test_cluster_storage.py": 47.17109420289855, - "unittest/disaggregated/test_coordinator_e2e.py": 72.26801550387597, - "unittest/disaggregated/test_coordinator_worker.py": 32.890360103626946, - "unittest/disaggregated/test_deepseek_v4_kv_transfer.py": 530.9268578431372, - "unittest/disaggregated/test_disagg_cluster_manager_worker.py": 55.52555365853659, - "unittest/disaggregated/test_disagg_openai_client.py": 33.3553485576923, - "unittest/disaggregated/test_disagg_utils.py": 23.560001321003963, - "unittest/disaggregated/test_disaggregated_params.py": 21.249914201183433, - "unittest/disaggregated/test_extractor.py": 26.12241229193342, - "unittest/disaggregated/test_kv_transfer.py": 607.4314255319149, - "unittest/disaggregated/test_kv_transfer_mp.py": 104.17107902735562, - "unittest/disaggregated/test_mamba_transfer.py": 33.60644819277109, - "unittest/disaggregated/test_messenger.py": 23.172675066312998, - "unittest/disaggregated/test_minimax_m3_kv_transfer.py": 337.1307835051547, - "unittest/disaggregated/test_openai_disagg_server.py": 27.25944660194175, - "unittest/disaggregated/test_openai_disagg_service.py": 33.72867386091127, - "unittest/disaggregated/test_openai_server_info.py": 28.47416666666667, - "unittest/disaggregated/test_peer.py": 23.180986541049798, - "unittest/disaggregated/test_perf_logger.py": 20.14458017492711, - "unittest/disaggregated/test_pool_matching.py": 19.757593406593404, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first and not ctx_first_sync\"": 442.66612371134016, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first_sync\"": 409.2202551020408, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first1\"": 446.7769587628866, - "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first2\"": 422.67712244897956, - "unittest/disaggregated/test_rank_info.py": 19.967412742382272, - "unittest/disaggregated/test_remoteDictionary.py": 28.488994572591587, - "unittest/disaggregated/test_request_id.py": 19.980398791540786, - "unittest/disaggregated/test_router.py": 25.549286479250334, - "unittest/disaggregated/test_sync_transfer_timing.py": 19.801176470588235, - "unittest/disaggregated/test_transceiver_bounded_polling.py": 20.033817204301073, - "unittest/dynamo": 30.93964938271605, - "unittest/executor/test_base_worker.py": 131.1828982630273, - "unittest/executor/test_fatal_error_health_check.py": 28.003123831775703, - "unittest/executor/test_ipc.py": 29.161146853146853, - "unittest/executor/test_multi_frontend_routing.py": 20.379634020618557, - "unittest/executor/test_postprocessor_hook.py": 29.60544927536232, - "unittest/executor/test_proxy_fast_death.py": 30.611908536585364, - "unittest/executor/test_proxy_postproc_terminate.py": 25.622395683453238, - "unittest/executor/test_rpc.py": 428.50881176470585, - "unittest/executor/test_stats_serializer.py": 22.270259154929576, - "unittest/inputs/test_chat_template_dispatch.py": 27.11566503667482, - "unittest/inputs/test_content_format.py": 25.79082608695652, - "unittest/inputs/test_multimodal.py": 25.693597087378638, - "unittest/inputs/test_multimodal_input_processor.py": 25.525548309178742, - "unittest/inputs/test_url_validation.py": 25.912739659367396, - "unittest/inputs/test_video_decode.py": 25.766543046357615, - "unittest/kv_cache_manager_v2_tests": 213.1598582781457, - "unittest/llmapi/apps/_test_openai_chat_multimodal.py::test_single_chat_session_image_embeds -m needs_l40s": 160.2151168224299, - "unittest/llmapi/apps/_test_openai_embeddings.py": 156.40381323877068, - "unittest/llmapi/apps/_test_openai_multi_gpu.py -m \"part0\"": 119.85816666666668, - "unittest/llmapi/apps/test_chat_utils.py": 27.616254716981132, - "unittest/llmapi/apps/test_disagg_serving_perf_metrics.py": 78.61523157894736, - "unittest/llmapi/apps/test_encode_batcher.py": 27.48177108433735, - "unittest/llmapi/apps/test_harmony_channel_validation.py": 30.0566327014218, - "unittest/llmapi/apps/test_tool_parsers.py": 29.678218009478673, - "unittest/llmapi/test_additional_model_outputs.py -m \"gpu1\"": 61.69495260663508, - "unittest/llmapi/test_async_llm.py -m \"gpu2\"": 83.43303092783505, - "unittest/llmapi/test_async_llm.py -m \"gpu4\"": 89.11570833333333, - "unittest/llmapi/test_async_llm.py -m \"not (gpu2 or gpu4)\"": 201.0308156424581, - "unittest/llmapi/test_deepseek_v4_tokenizer.py": 20.679311764705883, - "unittest/llmapi/test_embedding_arch_routing.py": 26.95822033898305, - "unittest/llmapi/test_executor.py": 27.13577777777778, - "unittest/llmapi/test_gc_utils.py": 32.136959427207636, - "unittest/llmapi/test_kv_cache_dtype_override.py": 27.04092523364486, - "unittest/llmapi/test_llm.py -m \"not part0\"": 214.32133333333334, - "unittest/llmapi/test_llm.py -m \"part0\"": 323.924, - "unittest/llmapi/test_llm_args.py": 131.87056698564595, - "unittest/llmapi/test_llm_encode.py": 61.49556812339331, - "unittest/llmapi/test_llm_encode_multi_item.py": 126.70357742782153, - "unittest/llmapi/test_llm_kv_cache_events.py::test_llm_api_attention_dp_kv_events": 53.872166666666665, - "unittest/llmapi/test_llm_multi_gpu_pytorch.py -m \"gpu4\"": 72.83206930693069, - "unittest/llmapi/test_llm_pytorch.py -m \"part0\"": 919.85638593482, - "unittest/llmapi/test_llm_pytorch.py -m \"part1\"": 258.6908440860215, - "unittest/llmapi/test_llm_pytorch.py -m \"part2\"": 516.3453256445048, - "unittest/llmapi/test_llm_pytorch.py -m \"part3\"": 280.88529755434786, - "unittest/llmapi/test_llm_pytorch.py::test_nemotron_nas_lora": 197.97414285714288, - "unittest/llmapi/test_llm_quant.py": 27.24870817843866, - "unittest/llmapi/test_llm_telemetry.py": 175.43786629526463, - "unittest/llmapi/test_llm_telemetry.py::TestTelemetryArchitectureExtraction": 72.02631630170316, - "unittest/llmapi/test_llm_telemetry.py::TestTelemetryPyTorchBackend": 71.87491931540343, - "unittest/llmapi/test_llm_utils.py": 31.80273798076923, - "unittest/llmapi/test_memory_profiling.py::test_pyexecutor_and_kvcache_share_execution_stream": 41.98156234096692, - "unittest/llmapi/test_mpi_session.py": 156.37610396039605, - "unittest/llmapi/test_mpi_session.py::test_llmapi_launch_multiple_tasks": 88.6355319148936, - "unittest/llmapi/test_reasoning_parser.py": 29.937814285714285, - "unittest/llmapi/test_request_priority.py": 25.679375296912113, - "unittest/llmapi/test_serialization.py": 25.607498806682578, - "unittest/llmapi/test_utils.py": 27.70747016706444, - "unittest/metrics/test_collector.py": 23.620726519337016, - "unittest/models/test_quant_config_utils.py": 25.616273809523808, - "unittest/others/test_convert_utils.py": 25.440990243902437, - "unittest/others/test_kv_cache_transceiver.py::test_async_transfer_keeps_llm_request_alive": 12.929715589541194, - "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mha]": 28.022305486900642, - "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mla]": 16.43778546712803, - "unittest/others/test_kv_cache_transceiver.py::test_cpp_nixl_sync_transfer_stress": 31.1141288945796, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[NIXL-mha-ctx_fp16_gen_fp16]": 24.296445485939813, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mha-ctx_fp16_gen_fp16]": 25.86616713091922, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mla-ctx_fp16_gen_fp16]": 24.195852528089887, - "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[UCX-mha-ctx_fp16_gen_fp16]": 12.57014578194376, - "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_silent_when_unset": 11.213987419832264, - "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_warns_once_per_request": 11.21132955106068, - "unittest/others/test_lora_manager.py": 36.074141463414634, - "unittest/others/test_lora_module_count.py": 27.08678883495146, - "unittest/others/test_time_breakdown.py": 27.776176904176904, - "unittest/others/test_tracing.py": 147.28475609756097, - "unittest/quantization/test_mode.py": 25.56721153846154, - "unittest/scaffolding": 446.7132156862745, - "unittest/scripts": 31.32473975903614, - "unittest/tools": 376.44609876543205, - "unittest/tools/test_layer_wise_benchmarks.py::test_deepseek_r1_ctx_dep[1]": 20.500902515723272, - "unittest/tools/test_layer_wise_benchmarks.py::test_nemotron_gen_dep[1]": 114.82735611510792, - "unittest/tools/test_layer_wise_benchmarks.py::test_performance_alignment[1]": 29.439571428571426, - "unittest/tools/test_layer_wise_benchmarks.py::test_qwen3_next_gen_tep[1]": 88.21061647058823, - "unittest/usage/test_collectors.py": 23.418028416779432, - "unittest/usage/test_config.py": 23.08291386271871, - "unittest/usage/test_e2e_capture.py": 23.17349802890933, - "unittest/usage/test_llmapi_config_telemetry_docs.py": 28.078541871921182, - "unittest/usage/test_opt_out.py": 23.03156282722513, - "unittest/usage/test_reporter.py": 26.090890645586295, - "unittest/usage/test_schema.py": 23.357160167130917, - "unittest/usage/test_transport.py": 29.417058265582654, - "unittest/utils/test_logger.py": 25.89497136038186, - "unittest/utils/test_util.py": 28.75711650485437, - "unittest/visual_gen/test_iteration_stats.py": 25.598708624708625, - "unittest/visual_gen/test_media_encoding.py": 25.9517546728972, - "unittest/visual_gen/test_output.py": 25.937927272727272, - "verl/test_verl_cases.py::test_async_generate": 424.8718181818182, - "verl/test_verl_cases.py::test_async_memory_management": 93.11290909090908, - "verl/test_verl_cases.py::test_init_without_device_mesh": 119.52972727272729, - "verl/test_verl_cases.py::test_inter_node_trtllm_rollout": 121.82945454545454, - "verl/test_verl_cases.py::test_make_async_request_get_method": 24.27409090909091, - "verl/test_verl_cases.py::test_make_async_request_http_error": 24.187363636363635, - "verl/test_verl_cases.py::test_make_async_request_max_attempts_exceeded": 24.261909090909093, - "verl/test_verl_cases.py::test_make_async_request_post_method": 25.87009090909091, - "verl/test_verl_cases.py::test_multimodal_different_image_sizes": 171.35136363636366, - "verl/test_verl_cases.py::test_multimodal_generate_with_image": 127.19154545454546, - "verl/test_verl_cases.py::test_multimodal_text_only_fallback": 115.52581818181818, - "verl/test_verl_cases.py::test_placement_group_multi_node_multi_replica": 21.758363636363637, - "verl/test_verl_cases.py::test_placement_group_multi_node_ray_resource_pool": 21.58818181818182, - "verl/test_verl_cases.py::test_placement_group_with_ray_resource_pool": 21.164454545454543, - "verl/test_verl_cases.py::test_placement_group_with_sub_ray_resource_pool": 21.24172727272727, - "verl/test_verl_cases.py::test_unimodal_batch_generate": 128.48545454545453, - "verl/test_verl_cases.py::test_unimodal_generate": 103.0079090909091, - "verl/test_verl_cases.py::test_wake_sleep_cycle": 106.03390909090909, - "visual_gen/test_visual_gen_benchmark.py::test_offline_benchmark": 74.4884, - "visual_gen/test_visual_gen_benchmark.py::test_online_benchmark[openai-videos]": 96.1846 + "accuracy/test_disaggregated_serving.py::TestDeepSeekR1::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 563.5671428571428, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV32Exp::test_auto_dtype[False]": 577.3825454545454, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=False]": 248.946875, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=0-overlap_scheduler=True]": 155.819, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=False]": 226.137625, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype[mtp_nextn=2-overlap_scheduler=True]": 155.884, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1dp2cp2]": 274.7764117647059, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[adp-mtp2]": 309.67263636363634, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_first[noadp-mtp0]": 459.758, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_gen_only_spec_dec": 174.414125, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=0]": 164.20842857142858, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[llguidance-mtp_nextn=2]": 373.866, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 141.08328571428572, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 176.04781981981984, + "accuracy/test_disaggregated_serving.py::TestDeepSeekV4Flash::test_auto_dtype": 653.3680224719101, + "accuracy/test_disaggregated_serving.py::TestGLM52NVFP4::test_nvfp4_nixl_python[cache_mgr_v1]": 1014.7549897959184, + "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 337.72714285714284, + "accuracy/test_disaggregated_serving.py::TestGPTOSS::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 431.009, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[False]": 250.61634653465347, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_auto_dtype[True]": 243.31274747474748, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v1]": 247.33391089108912, + "accuracy/test_disaggregated_serving.py::TestGemma3_1BInstruct::test_kv_cache_v2_nixl_python[cache_mgr_v2]": 423.1184242424243, + "accuracy/test_disaggregated_serving.py::TestKimiK25::test_nvfp4": 1168.2368000000001, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-False]": 205.6069387755102, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-False-True]": 205.10983333333334, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-False]": 213.14781632653063, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-False-True-True]": 213.95441836734696, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-False]": 164.90764646464646, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-False-True]": 162.11889795918367, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-False]": 176.48985294117648, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[False-True-True-True]": 175.97751515151515, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-False]": 204.73766336633665, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-False-True]": 202.2888383838384, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-False]": 216.0393039215686, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-False-True-True]": 214.77772549019608, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-False]": 163.57239603960394, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-False-True]": 161.92363725490196, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-False]": 176.5338673469388, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_auto_dtype[True-True-True-True]": 173.95371287128714, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_beam_search": 819.4241428571429, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=1-ctx_pp=2]": 158.725, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[GSM8K-gen_tp=2-ctx_pp=2]": 133.752, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[MMLU-gen_tp=1-ctx_pp=2]": 85.508, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_ctx_pp_gen_tp_asymmetric[MMLU-gen_tp=2-ctx_pp=2]": 85.066, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_eagle3[eagle3_one_model=True-overlap_scheduler=True]": 168.67477142857143, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_gen_only_spec_dec": 211.36951485148515, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 77.02336633663367, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 157.4601919191919, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_kv_cache_v2_nixl_python": 504.60869444444444, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[GSM8K]": 176.20839999999998, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_multi_instance[MMLU]": 108.71153, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp1pp2]": 198.364, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[GSM8K-tp2pp1]": 138.52641836734696, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp1pp2]": 86.348, + "accuracy/test_disaggregated_serving.py::TestLlama3_1_8BInstruct::test_tp_pp_symmetric[MMLU-tp2pp1]": 91.4269603960396, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=False]": 522.0529894736842, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=0-block_reuse=False-use_py_transceiver=True]": 626.4120731707317, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_auto_dtype[mtp_nextn=3-block_reuse=True-use_py_transceiver=False]": 476.09568085106383, + "accuracy/test_disaggregated_serving.py::TestNemotron3Super120B::test_ctx_dp2_gen_tp4": 589.0531538461539, + "accuracy/test_disaggregated_serving.py::TestQwen3NextInstruct::test_auto_dtype[use_py_transceiver=False]": 438.7, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-False]": 341.3867755102041, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[False-True]": 342.382137254902, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype[True-True]": 320.5173163265306, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2]": 296.2844343434343, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_chunked_prefill": 340.61881372549016, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first": 1049.90698, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_gen_first_kv_cache_v1": 129.3505, + "accuracy/test_disaggregated_serving.py::TestQwen3_8B::test_nixl_backend": 174.3157075471698, + "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_fp8]": 254.05641666666665, + "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[nemotron_nano_v3_omni_nvfp4]": 325.10448793565683, + "accuracy/test_epd_disagg_multimodal.py::TestVideoMMEEPD::test_disaggregated_videomme[qwen3vl_2b_instruct]": 181.72481818181816, + "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[no_overlap]": 66.90779939209726, + "accuracy/test_kv_pool_rebalance_accuracy.py::TestKvPoolRebalanceAccuracy::test_rebalance_matches_baseline[overlap]": 66.4708875, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-ep2]": 192.40157142857143, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b-tp2]": 294.607, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[120b]": 153.4064285714286, + "accuracy/test_llm_api_autodeploy.py::TestGPTOSS::test_mxfp4_gsm8k[20b]": 131.14983333333333, + "accuracy/test_llm_api_autodeploy.py::TestGemma4MoE::test_bf16": 1030.2359108910891, + "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma3n_e2b_it": 662.0174712990937, + "accuracy/test_llm_api_autodeploy.py::TestGemmaE2B::test_gemma4_e2b_it": 347.1145662650602, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_attention_dp[4]": 260.8819504950495, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[torch-True-1]": 928.2698, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[triton-False-1]": 297.40318934911244, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-1]": 151.84377401129942, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-False-4]": 130.9596782178218, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B::test_auto_dtype[trtllm-True-1]": 297.85671428571425, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[flashinfer]": 194.51192857142857, + "accuracy/test_llm_api_autodeploy.py::TestLlama3_1_8B_Instruct_Eagle3::test_eagle3_one_model[trtllm]": 159.54107079646016, + "accuracy/test_llm_api_autodeploy.py::TestMiniMaxM2::test_finegrained_fp8": 580.2765643564356, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[Qwen_QwQ-32B-False]": 269.0364285714286, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[deepseek-ai_DeepSeek-R1-0528-True]": 802.7421428571429, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[google_gemma-3-1b-it-False]": 58.604, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.1-8B-Instruct-False]": 47.470142857142854, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[meta-llama_Llama-3.3-70B-Instruct-False]": 140.37328571428571, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Codestral-22B-v0.1-False]": 87.76414285714286, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[mistralai_Ministral-8B-Instruct-2410-False]": 61.62542857142857, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_DeepSeek-R1-0528-NVFP4-v2-True]": 1478.76, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-FP8-True]": 107.06828571428571, + "accuracy/test_llm_api_autodeploy.py::TestModelRegistryAccuracy::test_autodeploy_from_registry[nvidia_Llama-3.1-8B-Instruct-NVFP4-True]": 125.27557425742575, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-flashinfer_ssm-False]": 438.54785714285714, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-False]": 356.0815714285714, + "accuracy/test_llm_api_autodeploy.py::TestNemotronH::test_auto_dtype[trtllm-triton_ssm-True]": 476.00142857142856, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-1-attn_dp_off-trtllm]": 1290.0525714285714, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 347.11914285714283, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-1-attn_dp_off-trtllm]": 389.7956233951498, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 260.72430348258706, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 219.20380198019802, + "accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 298.9254, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[bf16-4-attn_dp_off-trtllm]": 638.307, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_off-trtllm]": 476.6877425742574, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[fp8-4-attn_dp_on-trtllm]": 389.805, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-1-attn_dp_off-trtllm]": 742.2181908602151, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_accuracy[nvfp4-4-attn_dp_on-trtllm]": 546.3941666666666, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[bf16]": 243.7910835579515, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_functional_small[fp8]": 219.9212, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer]": 511.63059999999996, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm]": 392.782, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[fp8_ws4_80gb-trtllm]": 512.1759714285714, + "accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[nvfp4_ws4_80gb-trtllm]": 527.9766666666666, + "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-4]": 1914.121405940594, + "accuracy/test_llm_api_autodeploy.py::TestNemotronUltraV3::test_accuracy[nvfp4-8]": 2471.7431428571426, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[False]": 394.80692307692306, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[True]": 528.0706153846154, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_fp8[True]": 503.2507582260372, + "accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_nvfp4[True]": 415.859, + "accuracy/test_llm_api_autodeploy.py::TestQwen3_5_397B_MoE::test_nvfp4[8]": 1174.774, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[latency]": 2025.141625, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_fp8_blockscale[throughput]": 703.5015340136055, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency]": 814.2682857142856, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp]": 405.60157142857145, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_adp_lmtp_tp4]": 750.1278415841583, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen]": 803.0506666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[latency_trtllmgen_adp_lmtp]": 594.0771666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput]": 484.942600660066, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_bs8_mtp]": 616.6849142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_mtp]": 431.8518361774744, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp4]": 786.3406, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_multi_gpus[throughput_tp8]": 425.4071666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_dsa_host_cache_offload[host_cache_offload]": 675.9716666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline]": 708.164, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[baseline_mtp1]": 873.1390857142858, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_attn_multi_gpus": 620.9094285714285, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline]": 385.06701923076923, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline_mtp1]": 403.80720952380955, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[baseline]": 350.41063636363634, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3::test_skip_softmax_attention_multi_gpus[target_sparsity_0.9]": 1907.8475841584159, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 110.22128571428571, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 95.23816666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 112.59433333333332, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 117.33757142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 108.42275, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 96.67157142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 99.13671428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 104.493, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 136.719875, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 135.1635, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 93.43057142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 96.45733333333332, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 115.44271428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 108.53066666666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 173.38052512562814, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 240.23762278481013, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 89.461, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 98.76371428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=False]": 109.929625, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True-enable_chunked_prefill=False-v2_kv_cache=True]": 116.24385714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 87.24416666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 89.3735, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 171.456, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 178.44, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 89.22883333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=True]": 93.58716666666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=False-v2_kv_cache=False]": 233.06353576437587, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=False]": 490.6032997416021, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-enable_chunked_prefill=True-v2_kv_cache=True]": 672.0208601583113, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_2_model_mtp": 81.99830310880829, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 74.9602857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 157.13757142857142, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 50.04, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 53.55142857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.368857142857145, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 52.22757142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.32028571428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 43.58842857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 124.31714285714285, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 37.46957142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[ep4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 286.8615555555556, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[pp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 361.54089473684206, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 51.500142857142855, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 47.665571428571425, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.233, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 49.38442857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 50.68985714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 241.726, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 49.91157142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 60.44242857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 72.28114285714285, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp2pp2-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 185.46715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 54.29342857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 150.5922857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 44.819714285714284, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 53.9998, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 95.5758, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 45.01414285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 54.855714285714285, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 46.03628571428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 70.931, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 49.37942857142857, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 69.58571428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 177.16992307692306, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 39.854571428571425, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus[tp4-mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 64.18114285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=0]": 110.4788, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_kv_cache_aware_routing[mtp_nextn=2]": 65.1326, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=0]": 57.580400000000004, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[ep4-mtp_nextn=2]": 68.72460000000001, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=0]": 73.8476, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_4gpus_python_scheduler[tp4-mtp_nextn=2]": 74.2298, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=False]": 204.60704605263157, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_flashinfer[enable_chunked_prefill=True]": 317.57645359477124, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 242.8417142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 228.30883333333335, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=0-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 1257.124142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=False-cuda_graph=False-overlap_scheduler=False-enable_chunked_prefill=False]": 282.825, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=False]": 334.87314285714285, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_bfloat16_python_scheduler[mtp_nextn=2-attention_dp=True-cuda_graph=True-overlap_scheduler=True-enable_chunked_prefill=True]": 826.2921666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=fp8-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 450.24012227074235, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=False-fp8kv=False-overlap_scheduler=True]": 251.18992151162792, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=none-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 233.96449426934097, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=False-overlap_scheduler=True]": 228.73876377952757, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_chunked_prefill[quant_dtype=nvfp4-kv_cache_reuse=True-fp8kv=True-overlap_scheduler=True]": 293.3109759358289, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_dummy_load_format": 185.64599129172714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 125.61314285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 141.97885714285712, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 92.35671428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 100.544, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.229, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 114.00025, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.495125, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 202.64102366863904, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=disable-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 100.6405, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.03385714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 95.19957142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 117.24828571428571, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.07571428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 96.91071428571429, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=eagle-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 203.96484821428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales[mtp=vanilla-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 100.222, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 128.5577142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 55.65114285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 48.82, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 132.92722115384615, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 89.69357142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.26528571428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 63.23810576923076, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[ep4-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 61.52645192307692, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[pp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 58.106, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[pp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 60.694, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 94.36728571428571, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 47.680571428571426, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 43.60585714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 113.014, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 50.377, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 73.366, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 51.053, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 50.206857142857146, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 49.96328571428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp2pp2-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 69.58657142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 65.82942857142856, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 76.64985714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 59.55085714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 59.55585714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.456857142857146, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True-sampler_async_worker=False]": 81.1257142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 53.917857142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=True]": 68.30914285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 130.82932673267325, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 61.337153846153846, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 58.744692307692304, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 110.86007920792079, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=True]": 114.5420297029703, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=False]": 48.49171428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False-sampler_async_worker=True]": 58.338857142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 50.66985714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-sampler_async_worker=False]": 49.70885714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_4gpus[tp4-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True-sampler_async_worker=False]": 129.97968316831683, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=0]": 65.82992835820896, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding[mtp_nextn=2]": 146.14830721003133, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=0]": 51.27741747572816, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_fp8_block_scales_cuda_graph_padding_4gpus[attention_dp=True-mtp_nextn=2]": 58.42619417475728, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=0]": 74.263375, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding[xgrammar-mtp_nextn=2]": 229.74792266666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=0]": 106.62016666666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_guided_decoding_4gpus[xgrammar-mtp_nextn=2]": 64.91603883495145, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=fp8-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 142.49904672897196, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_no_kv_cache_reuse[quant_dtype=none-mtp_nextn=2-fp8kv=False-attention_dp=True-cuda_graph=True-overlap_scheduler=True]": 257.45325974025974, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 197.461, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTEDSL-mtp_nextn=2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 238.9005, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 129.17557142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 144.1535714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=True-torch_compile=False]": 118.58166666666668, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 125.60383333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 121.43585714285715, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-torch_compile=True]": 137.706, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 126.48357142857144, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=True]": 136.42479999999998, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=0-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 118.6963217158177, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 168.66416666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False]": 176.74114285714288, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=CUTLASS-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 315.0251083123425, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 123.97833333333332, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=0-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 94.24726816380449, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=False-attention_dp=False-cuda_graph=True-overlap_scheduler=False-torch_compile=False]": 198.32728571428572, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4[moe_backend=TRTLLM-mtp_nextn=2-fp8kv=True-attention_dp=False-cuda_graph=True-overlap_scheduler=True-torch_compile=False]": 206.059, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 188.77516666666665, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 357.6437142857143, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 413.533, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 235.61871428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 142.03583333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 204.68985714285714, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-pp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 264.8346666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 229.01957142857142, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTEDSL-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 369.8146712962963, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 159.7267619047619, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 147.81376923076922, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 84.19792307692308, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 100.28721428571428, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 214.388625, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 125.55446153846154, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-pp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 59.623, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 174.7900666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 169.331, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 72.06583333333333, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 134.51646666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 87.48892592592593, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 151.09442, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=True]": 159.31716666666665, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 95.5052197309417, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 113.31939285714286, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 155.69638461538463, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=False]": 145.39335, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 146.07366666666667, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 130.40482805429863, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-low_precision_combine=False-torch_compile=True]": 85.24828571428571, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=False-attention_dp=True-cuda_graph=False-overlap_scheduler=False-low_precision_combine=True-torch_compile=False]": 179.83242857142858, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 183.90396666666666, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=0-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=True-torch_compile=False]": 77.26146728971963, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-ep4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 261.2648656716418, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=TRTLLM-mtp_nextn=2-tp4-fp8kv=True-attention_dp=True-cuda_graph=True-overlap_scheduler=True-low_precision_combine=False-torch_compile=False]": 114.26407692307693, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus_online_eplb[fp8kv=True-moe_backend=TRTLLM]": 188.3487525773196, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_batch_waiting[batch_wait_timeout_iters=10-batch_wait_max_tokens_ratio=1.0-mtp_nextn=0-fp8kv=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False-v2_kv_cache=True]": 222.85041397849463, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_auto_dtype": 790.9375, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=MEGAMOE_DEEPGEMM]": 559.4773168316832, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Flash::test_nvfp4_4gpus_static_eplb[moe_backend=TRTLLM]": 327.3962673267327, + "accuracy/test_llm_api_pytorch.py::TestDeepSeekV4Pro::test_gsm8k_full_accuracy": 1220.98871875, + "accuracy/test_llm_api_pytorch.py::TestGLM52::test_nvfp4_mtp_index_share[tp_size=8-ep_size=8]": 801.905, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dflash": 606.7177058823529, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_dummy_load_format": 82.15868983957219, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[cutlass-one_model-overlap_scheduler]": 1810.228, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_2gpus[triton-one_model-overlap_scheduler]": 1204.1007378640775, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-no_overlap_scheduler]": 597.1911333333333, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-cutlass-one_model-overlap_scheduler]": 446.03928155339804, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-no_overlap_scheduler]": 229.9775, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v1_kv_cache-trtllm-one_model-overlap_scheduler]": 411.70846521739134, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-no_overlap_scheduler]": 474.76293333333336, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-cutlass-one_model-overlap_scheduler]": 644.7451866028708, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-no_overlap_scheduler]": 397.8118333333333, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_4gpus[v2_kv_cache-trtllm-one_model-overlap_scheduler]": 505.294358490566, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-one_model]": 259.8957142857143, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v1_kv_cache-two_model]": 324.8745714285714, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_eagle3_vswa_reuse_4gpus[v2_kv_cache-one_model]": 708.3871666666666, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-cutlass-auto]": 238.89962115127173, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-auto]": 283.65757142857143, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v1_kv_cache-True-True-trtllm-fp8]": 199.45164247311828, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-cutlass-auto]": 173.85969541778977, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-auto]": 201.94083333333333, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_1gpu[v2_kv_cache-True-True-trtllm-fp8]": 315.4943882978723, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-triton-auto]": 1349.77619, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-auto]": 536.0184, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-dp4-trtllm-fp8]": 459.6693106796116, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-triton-auto]": 512.76651, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-ep4-trtllm-auto]": 446.99943925233646, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-triton-auto]": 456.39875, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v1_kv_cache-tp4-trtllm-auto]": 544.3978755980862, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-triton-auto]": 844.6485600000001, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-auto]": 642.8702, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-trtllm-fp8]": 573.480285, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-triton-auto]": 632.22521, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-auto]": 478.0725, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-trtllm-fp8]": 326.0081666666667, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-tp4-trtllm-auto]": 536.486418079096, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache_no_reuse-tp4-cutlass-auto]": 412.47744, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus_online_eplb[fp8]": 276.85751871657754, + "accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]": 331.16407, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype": 194.39304497751124, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_reuse": 119.92895819935691, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_chunked_prefill_without_reuse": 130.76036677115988, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse": 47.05186774193549, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_disable_overlap_scheduler": 46.59689393939394, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_no_partial_reuse": 61.084156050955414, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_low_memory_available_partial_reuse": 62.60442056074766, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_reuse_partial_reuse": 44.13613738019169, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse": 43.938835913312694, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_disable_overlap_scheduler": 48.01080909090909, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype_vswa_without_reuse_low_memory_available": 56.98198746081505, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=False]": 180.3642857142857, + "accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_fp8_prequantized[torch_compile=True]": 207.9544285714286, + "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_auto_dtype": 895.99875, + "accuracy/test_llm_api_pytorch.py::TestGemma3_27BInstruct::test_fp8_prequantized": 418.2255, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8]": 439.1708333333333, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8]": 679.1593333333334, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8]": 974.601006779661, + "accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8_attn_dp]": 783.5843333333333, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_bf16": 336.4756683291771, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_fp8": 246.60832780082987, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS::test_nvfp4": 206.84562613195345, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_bf16_dflash": 310.8462520547945, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_fp8_dflash": 304.4109135802469, + "accuracy/test_llm_api_pytorch.py::TestLagunaXS_2_1::test_nvfp4_dflash": 303.8202086720867, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4": 147.18037917222964, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_4]": 117.7441439790576, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B::test_nvfp4_streaming[stream_interval_64]": 124.89551358695653, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=False]": 96.55514285714285, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=FLASHINFER-torch_compile=True]": 128.05185714285716, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=False]": 78.98383333333332, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16[attn_backend=TRTLLM-torch_compile=True]": 113.277, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=FLASHINFER-torch_compile=False]": 119.661, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=FLASHINFER-torch_compile=False]": 39.89528571428572, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=False]": 38.87285714285714, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=TRTLLM-torch_compile=True]": 45.632714285714286, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=False]": 94.90157575757576, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=FLASHINFER-torch_compile=True]": 68.93085714285715, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=TRTLLM-torch_compile=True]": 106.232, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=FLASHINFER]": 153.13166666666666, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=False-attn_backend=TRTLLM]": 48.5009577039275, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=FLASHINFER]": 164.974125, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_chunked_prefill[use_temperature=True-attn_backend=TRTLLM]": 59.59303960396039, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dflash": 165.76832544378698, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_dummy_load_format": 18.441307692307692, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=False-eagle3_one_model=True-overlap_scheduler=True]": 976.3793964497042, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_eagle3[sampler_async_worker=True-eagle3_one_model=True-overlap_scheduler=True]": 971.4979968652037, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 67.56983333333334, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 109.339625, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 53.60133641975309, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 81.8483850931677, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 79.50657142857143, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 97.091, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 46.40235435435436, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8[fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 70.09814423076924, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[pp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 89.04066666666667, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 40.44528571428572, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 130.926, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 38.101142857142854, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp2pp2-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 33.152, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=False]": 113.30497727272729, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=FLASHINFER-torch_compile=True]": 72.48057142857144, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=False]": 126.10539603960396, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=False-attn_backend=TRTLLM-torch_compile=True]": 87.13407272727272, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=False]": 39.70928571428572, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=FLASHINFER-torch_compile=True]": 130.6536153846154, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=False]": 63.65173205741627, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_fp8_4gpus[tp4-fp8kv=True-attn_backend=TRTLLM-torch_compile=True]": 81.74710891089109, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[llguidance]": 46.254285714285714, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding[xgrammar]": 48.451914373088684, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[llguidance]": 47.952, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_4gpus[xgrammar]": 49.702, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[llguidance-eagle3_one_model=True]": 57.91375, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_eagle3[xgrammar-eagle3_one_model=True]": 45.978490797546016, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[llguidance]": 44.04633333333334, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_guided_decoding_with_ngram[xgrammar]": 43.43625, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=False]": 62.13226470588235, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=False-attn_backend=TRTLLM-torch_compile=True]": 78.61358743169399, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_nvfp4_kv[v2_kv_cache=True-attn_backend=TRTLLM-torch_compile=True]": 233.3722371967655, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=False]": 780.0762724458203, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_pard[overlap_scheduler=True]": 631.2467230769231, + "accuracy/test_llm_api_pytorch.py::TestLlama3_1_8B_Instruct_RocketKV::test_auto_dtype": 930.7815714285715, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp4_tp2pp2[torch_compile=True-enable_gemm_allreduce_fusion=False]": 827.042, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=False]": 671.1513591160221, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_fp8_tp4[torch_compile=True]": 653.0381851851852, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=False]": 743.1877736842105, + "accuracy/test_llm_api_pytorch.py::TestLlama3_3_70BInstruct::test_nvfp4_tp4[torch_compile=True]": 747.0298399999999, + "accuracy/test_llm_api_pytorch.py::TestMiniMaxM3::test_nvfp4[use_msa=True]": 1015.7231386138614, + "accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm]": 529.4998, + "accuracy/test_llm_api_pytorch.py::TestMistralLarge3_675B::test_nvfp4_4gpus[latency_moe_trtllm_eagle]": 764.3608, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_fp8": 301.294, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Nano::test_nvfp4_marlin_multi_gpus[tp_size=8]": 409.00075, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-cpp_mamba_cache]": 376.7685, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_off-python_mamba_cache]": 308.124, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-cpp_mamba_cache]": 214.224, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_fp8_4gpus[attention_dp_on-python_mamba_cache]": 215.894, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar": 383.92656190476185, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpu_mtp_ar_custom_op": 251.13214285714287, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_OFF]": 805.9374, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[DEP4_MTP_ON]": 499.62153535353536, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_block_reuse[TEP4]": 782.1175999999999, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_hopper_w4a16": 535.0051428571428, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_online_eplb[moe_backend=TRTLLM]": 485.27279999999996, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 394.49414285714283, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_8gpus_mtp": 715.7141650485437, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_adp_4gpus[mtp_nextn=3]": 948.6714285714286, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Super::test_nvfp4_marlin_multi_gpus[tp_size=8]": 448.4485, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_block_reuse[ADP4_MTP]": 1103.1094324324324, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 994.767, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_4gpus_static_eplb[moe_backend=CUTLASS]": 511.1553333333333, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_off-trtllm]": 555.4016868686869, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_8gpus[attention_dp_on-cutedsl]": 673.2446534653466, + "accuracy/test_llm_api_pytorch.py::TestNemotronV3Ultra::test_nvfp4_marlin_8gpus": 2393.7385, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_2gpu_mtp_ar": 296.3352, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[dep4]": 403.4035, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_bf16_4gpu[tep4]": 594.184, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-cutlass]": 310.09275, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[dep4-trtllm]": 288.0665, + "accuracy/test_llm_api_pytorch.py::TestQwen3NextInstruct::test_nvfp4[tp1_block_reuse-cutlass]": 330.1165, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_cutlass]": 221.4197142857143, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency_moe_trtllm_attention_dp]": 265.43, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_cutlass]": 1021.0351428571429, + "accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4_4gpus[latency_moe_trtllm_eagle3]": 325.0231538461539, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_dummy_load_format": 41.01788461538462, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=False]": 156.23925, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8[latency-torch_compile=True]": 189.51606918238994, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=False]": 122.25, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_fp8_block_scales[latency-torch_compile=True]": 143.73514285714288, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]": 122.8515, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=True]": 177.88014285714286, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=True]": 124.60616666666667, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=False]": 212.62992307692306, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_cutlass-torch_compile=True]": 218.3575, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=False]": 203.124, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[latency_moe_trtllm-torch_compile=True]": 169.98084864864865, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=False]": 258.0261875, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_cutlass-torch_compile=True]": 286.8506956521739, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=False]": 201.9002142857143, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]": 211.51953271028037, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a16_mxfp4[latency-TRTLLM]": 77.09778494623656, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-CUTLASS]": 85.84242857142857, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[fp8-latency-TRTLLM]": 90.60216981132075, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-CUTLASS]": 80.64685714285714, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_w4a8_mxfp4[mxfp8-latency-TRTLLM]": 76.41928421052631, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention[target_sparsity_0.9-fp8kv=True]": 813.8270840108402, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=False]": 709.772, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.5-fp8kv=True]": 554.5961538461539, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=False]": 581.32, + "accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B_Instruct_2507::test_skip_softmax_attention_4gpus[target_sparsity_0.9-fp8kv=True]": 508.1515714285714, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-CUTLASS]": 275.259, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp1-TRTLLM]": 225.91433333333333, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-CUTLASS]": 235.185, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16[tp2-TRTLLM]": 309.260198019802, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_bf16_mtp": 342.53534065934065, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=False]": 438.6115, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_35B_A3B::test_fp8[enable_block_reuse=True]": 646.5637106918239, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_cutedsl]": 617.0652068965518, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[adp4_trtllm]": 461.029, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_block_reuse]": 632.7510101010101, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4[tep4_trtllm]": 534.2323333333334, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_online_eplb[moe_backend=CUTEDSL]": 955.435089201878, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_4gpus_static_eplb[moe_backend=CUTEDSL]": 662.74585, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_397B_A17B::test_nvfp4_mtp3_gdn_replay_tep4": 526.8681475409836, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_bf16": 166.71264935064934, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_dflash": 214.72679870129872, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8": 207.41428353658537, + "accuracy/test_llm_api_pytorch.py::TestQwen3_5_4B::test_fp8_piecewise_cuda_graph": 198.19608035714288, + "accuracy/test_llm_api_pytorch.py::TestQwen3_6_35B_A3B::test_nvfp4[TRTLLM]": 270.3366730245232, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dflash": 143.1653267973856, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_dummy_load_format": 44.24248447204969, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=False]": 100.20931804281345, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=False-enable_draft_len_schedule=True]": 107.118, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=False-enable_max_concurrency=True-enable_draft_len_schedule=False]": 105.864125, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_eagle3[eagle3_one_model=True-enable_chunked_prefill=True-enable_max_concurrency=False-enable_draft_len_schedule=False]": 194.1835, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales[latency]": 217.11881497480204, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_fp8_block_scales_early_first_token_response": 135.53371608832808, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[fp8-latency]": 38.061456464379944, + "accuracy/test_llm_api_pytorch.py::TestQwen3_8B::test_w4a8_mxfp4[mxfp8-latency]": 48.70666666666666, + "accuracy/test_llm_api_pytorch.py::TestSeedOss_36B::test_auto_dtype": 1182.1868333333332, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_cuda_graph_matches_eager_logits[tinyllama-1.1b]": 14.049864583333333, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[gemma-3-1b]": 18.696708333333333, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[phi-4-mini]": 17.96046875, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen2-7b]": 19.755377604166668, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[qwen3-0.6b]": 8.566161458333333, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[starcoder2-3b]": 24.163671875, + "accuracy/test_llm_api_pytorch_encode.py::TestDecoderEncode::test_decoder_encode_matches_huggingface[tinyllama-1.1b]": 6.4027109375, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_cuda_graph_matches_eager_logits[bert-yelp]": 6.059911458333333, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-cuda_graph]": 4.2082109375, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_classification[bert-yelp-eager]": 4.435528645833333, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_encoder_encode_matches_huggingface_per_token_reward[qwen2.5-prm-7b]": 22.102708333333332, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-0.6b]": 11.882435443037975, + "accuracy/test_llm_api_pytorch_encode.py::TestEncoderEncode::test_qwen3_text_embedding_matches_huggingface[qwen3-embedding-8b]": 25.415469696969698, + "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_fp8_prequantized": 328.788006006006, + "accuracy/test_llm_api_pytorch_multimodal.py::TestGemma4_26B_A4B::test_nvfp4": 345.3036666666667, + "accuracy/test_llm_api_pytorch_multimodal.py::TestMistralSmall24B::test_auto_dtype[forced_chunked_prefill]": 171.36042857142857, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[bf16]": 418.94666666666666, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8]": 651.2778228228228, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[fp8_mmmu_encoder_cuda_graph]": 326.4243922651934, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNanoV3Omni::test_auto_dtype[nvfp4]": 467.0641428571428, + "accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype[forced_chunked_prefill]": 277.644, + "accuracy/test_llm_api_pytorch_multimodal.py::TestQwen3VL::test_auto_dtype[forced_chunked_prefill]": 155.48857142857142, + "accuracy/test_llm_api_pytorch_ray.py::TestLlama3_1_8BInstruct::test_pp2_ray": 72.727, + "cpp/test_multi_gpu.py::test_cache_transceiver[2proc-ucx_kvcache-90]": 185.5999120879121, + "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-nixl_kvcache-90]": 248.42608791208792, + "cpp/test_multi_gpu.py::test_cache_transceiver[8proc-ucx_kvcache-90]": 317.74205555555557, + "cpp/test_multi_gpu.py::test_fused_gemm_allreduce[4proc-90]": 21.271241758241757, + "cpp/test_multi_gpu.py::test_mpi_utils[90]": 231.0067802197802, + "cpp/test_multi_gpu.py::test_user_buffer[2proc-90]": 4.18345054945055, + "cpp/test_unit_tests.py::test_unit_tests[batch_manager-80]": 561.8357163323782, + "cpp/test_unit_tests.py::test_unit_tests[common-80]": 225.72357142857143, + "cpp/test_unit_tests.py::test_unit_tests[common-90]": 224.40657185628743, + "cpp/test_unit_tests.py::test_unit_tests[executor-80]": 130.4267142857143, + "cpp/test_unit_tests.py::test_unit_tests[kernels-80]": 943.2667142857143, + "cpp/test_unit_tests.py::test_unit_tests[kernels-90]": 731.6601886227545, + "cpp/test_unit_tests.py::test_unit_tests[layers-80]": 918.556, + "cpp/test_unit_tests.py::test_unit_tests[layers-90]": 794.5549492537313, + "cpp/test_unit_tests.py::test_unit_tests[runtime-80]": 678.604, + "cpp/test_unit_tests.py::test_unit_tests[thop-80]": 3.1942857142857144, + "cpp/test_unit_tests.py::test_unit_tests[thop-90]": 3.119434131736527, + "disaggregated/test_ad_disagg.py::test_async_eagle3_full_model_handoff": 97.08721782178218, + "disaggregated/test_ad_disagg.py::test_async_generation_matches_aggregate": 86.58474257425743, + "disaggregated/test_ad_disagg.py::test_async_generation_no_overlap_matches_aggregate": 84.61831683168317, + "disaggregated/test_ad_disagg.py::test_async_sharded_generation_handoff": 97.06769, + "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[deepseek_v3_mla]": 157.40455151515152, + "disaggregated/test_ad_disagg.py::test_chunked_prefill_handoff[tinyllama]": 96.72484194528876, + "disaggregated/test_ad_disagg.py::test_disaggregated_logits[deepseek_v3_mla]": 56.42044545454546, + "disaggregated/test_ad_disagg.py::test_disaggregated_logits[tinyllama]": 34.52554819277108, + "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[deepseek_v3_mla]": 141.5973172205438, + "disaggregated/test_ad_disagg.py::test_reduced_layer_handoff_matches_aggregate[tinyllama]": 96.18908459214501, + "disaggregated/test_ad_disagg.py::test_tinyllama_batch_handoff_semantic_slots": 76.65452567975831, + "disaggregated/test_ad_disagg_trtllm_serve.py::test_openai_completion": 61.01671428571428, + "disaggregated/test_auto_scaling.py::test_disagg_server_restart[etcd-round_robin]": 104.70482000000001, + "disaggregated/test_auto_scaling.py::test_disagg_server_restart[http-round_robin]": 133.51976530612245, + "disaggregated/test_auto_scaling.py::test_minimal_instances[etcd-round_robin]": 103.74983673469387, + "disaggregated/test_auto_scaling.py::test_minimal_instances[http-round_robin]": 101.29590816326531, + "disaggregated/test_auto_scaling.py::test_service_discovery[etcd-round_robin]": 54.902530612244895, + "disaggregated/test_auto_scaling.py::test_service_discovery[http-kv_cache_aware]": 51.200849999999996, + "disaggregated/test_auto_scaling.py::test_service_discovery[http-round_robin]": 50.95837, + "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-load_balancing]": 205.29111224489796, + "disaggregated/test_auto_scaling.py::test_worker_restart[etcd-round_robin]": 202.5159595959596, + "disaggregated/test_auto_scaling.py::test_worker_restart[http-load_balancing]": 201.06501999999998, + "disaggregated/test_disaggregated.py::test_disaggregated_benchmark_gen_only_insufficient_kv[TinyLlama-1.1B-Chat-v1.0]": 75.97949742268042, + "disaggregated/test_disaggregated.py::test_disaggregated_cache_aware_balance[TinyLlama-1.1B-Chat-v1.0]": 132.24700515463917, + "disaggregated/test_disaggregated.py::test_disaggregated_chat_completion_tool_calls[TinyLlama-1.1B-Chat-v1.0]": 71.68480506329114, + "disaggregated/test_disaggregated.py::test_disaggregated_conditional[TinyLlama-1.1B-Chat-v1.0]": 92.38360050251256, + "disaggregated/test_disaggregated.py::test_disaggregated_ctxpp2_genpp2[TinyLlama-1.1B-Chat-v1.0]": 71.839, + "disaggregated/test_disaggregated.py::test_disaggregated_ctxpp2_gentp2[TinyLlama-1.1B-Chat-v1.0]": 71.719, + "disaggregated/test_disaggregated.py::test_disaggregated_ctxtp2_genpp2[TinyLlama-1.1B-Chat-v1.0]": 69.536, + "disaggregated/test_disaggregated.py::test_disaggregated_cuda_graph[TinyLlama-1.1B-Chat-v1.0]": 80.77402295918367, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_conditional_v2[DeepSeek-V3-Lite-bf16]": 139.9444081632653, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_empty_batch[DeepSeek-V3-Lite-bf16]": 190.4600707070707, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 128.8671616161616, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 121.09470707070707, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8]": 118.99839393939394, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 99.04030303030302, + "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 136.5389226006192, + "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 72.53091237113402, + "disaggregated/test_disaggregated.py::test_disaggregated_gpt_oss_120b_harmony[gpt_oss/gpt-oss-120b]": 180.8998811881188, + "disaggregated/test_disaggregated.py::test_disaggregated_kv_cache_time_output[TinyLlama-1.1B-Chat-v1.0]": 92.5423762886598, + "disaggregated/test_disaggregated.py::test_disaggregated_load_balance[TinyLlama-1.1B-Chat-v1.0]": 83.46904984423675, + "disaggregated/test_disaggregated.py::test_disaggregated_mixed[TinyLlama-1.1B-Chat-v1.0]": 137.4564641025641, + "disaggregated/test_disaggregated.py::test_disaggregated_mixed_stress_test[req120-conc64-qwen3_32b_fp8_mixed_stress]": 199.7655, + "disaggregated/test_disaggregated.py::test_disaggregated_multi_gpu[TinyLlama-1.1B-Chat-v1.0]": 72.94412244897958, + "disaggregated/test_disaggregated.py::test_disaggregated_ngram[TinyLlama-1.1B-Chat-v1.0]": 87.9892202072539, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap[TinyLlama-1.1B-Chat-v1.0]": 100.85853292181069, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python[TinyLlama-1.1B-Chat-v1.0]": 80.67388775510204, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_bounce[TinyLlama-1.1B-Chat-v1.0]": 107.13556521739132, + "disaggregated/test_disaggregated.py::test_disaggregated_overlap_transceiver_runtime_python_fabric_memory[TinyLlama-1.1B-Chat-v1.0]": 104.9951875, + "disaggregated/test_disaggregated.py::test_disaggregated_perf_metrics[TinyLlama-1.1B-Chat-v1.0]": 90.53827720207254, + "disaggregated/test_disaggregated.py::test_disaggregated_python_transceiver_host_offload[TinyLlama-1.1B-Chat-v1.0]": 72.30680000000001, + "disaggregated/test_disaggregated.py::test_disaggregated_qwen3_32b_fp8[Qwen3/Qwen3-32B-FP8]": 146.544875, + "disaggregated/test_disaggregated.py::test_disaggregated_single_gpu[TinyLlama-1.1B-Chat-v1.0]": 86.30057989690722, + "disaggregated/test_disaggregated.py::test_disaggregated_tinyllama_multi_orchestrator[TinyLlama-1.1B-Chat-v1.0]": 68.38216216216216, + "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer[False-TinyLlama-1.1B-Chat-v1.0]": 57.86059788359788, + "disaggregated/test_disaggregated_single_gpu.py::test_arbitrary_kv_cache_transfer_missing_blocks[False-TinyLlama-1.1B-Chat-v1.0]": 60.099248, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_cancel_gen_requests[TinyLlama-1.1B-Chat-v1.0]": 34.60525471698113, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_llama_context_capacity[False-False-DeepSeek-V3-Lite-fp8/fp8]": 160.16489374999998, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[False-TinyLlama-1.1B-Chat-v1.0]": 49.2354213740458, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logits[True-TinyLlama-1.1B-Chat-v1.0]": 40.8620374251497, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[False-TinyLlama-1.1B-Chat-v1.0]": 47.87118947368421, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_logprobs[True-TinyLlama-1.1B-Chat-v1.0]": 45.21462158054712, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-False-DeepSeek-V3-Lite-fp8/fp8]": 99.36854098360655, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[False-True-DeepSeek-V3-Lite-fp8/fp8]": 78.65977846153847, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-False-DeepSeek-V3-Lite-fp8/fp8]": 127.02941666666668, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_deepseek[True-True-DeepSeek-V3-Lite-fp8/fp8]": 114.31295031055902, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-False-TinyLlama-1.1B-Chat-v1.0]": 44.86971506105834, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[False-True-TinyLlama-1.1B-Chat-v1.0]": 48.939882653061225, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-False-TinyLlama-1.1B-Chat-v1.0]": 56.65890439276486, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_llama[True-True-TinyLlama-1.1B-Chat-v1.0]": 58.08935368956743, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-False-Qwen3-8B-FP8]": 77.68070953101362, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[False-True-Qwen3-8B-FP8]": 70.43922955974844, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-False-Qwen3-8B-FP8]": 65.3021451104101, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_simple_qwen3[True-True-Qwen3-8B-FP8]": 68.98847094801222, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[False-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 56.87129813664596, + "disaggregated/test_disaggregated_single_gpu.py::test_disaggregated_spec_dec_batch_slot_limit[True-False-EAGLE3-LLaMA3.1-Instruct-8B-Llama-3.1-8B-Instruct]": 69.73002752293579, + "disaggregated/test_workers.py::test_workers_conditional_disaggregation_deepseek_v3_lite_bf16[DeepSeek-V3-Lite-bf16]": 145.47861616161617, + "disaggregated/test_workers.py::test_workers_kv_cache_aware_router[TinyLlama-1.1B-Chat-v1.0]": 137.83586597938142, + "disaggregated/test_workers.py::test_workers_kv_cache_events[TinyLlama-1.1B-Chat-v1.0]": 78.49960406091371, + "examples/test_ad_guided_decoding.py::test_autodeploy_guided_decoding_main_json": 108.45550755287009, + "examples/test_ad_speculative_decoding.py::test_autodeploy_eagle3_one_model_acceptance_rate[flashinfer-torch-simple]": 102.75090207715134, + "examples/test_ad_speculative_decoding.py::test_eagle_model_with_weights": 2.719433234421365, + "examples/test_ad_speculative_decoding.py::test_nemotron_mtp_model_with_weights": 799.8663165680474, + "examples/test_deepseek_v4_pro.py::test_short_token_boundary_smoke": 954.67404, + "examples/test_llm_api_with_mpi.py::test_llm_api_single_gpu_with_mpirun[TinyLlama-1.1B-Chat-v1.0]": 47.71125796812749, + "examples/test_ray.py::test_llm_inference_async_ray": 126.3797485380117, + "examples/test_ray.py::test_llm_inference_distributed_ray[pp2]": 112.781, + "examples/test_ray.py::test_llm_inference_distributed_ray[tep2]": 158.26951485148516, + "examples/test_ray.py::test_llm_inference_distributed_ray[tp2]": 65.46412871287129, + "examples/test_ray.py::test_llm_inference_distributed_ray[tp2pp2]": 196.643, + "examples/visual_gen/test_visual_gen.py::test_cosmos3_example": 286.4563489583333, + "examples/visual_gen/test_visual_gen.py::test_flux1_example": 238.85635549872123, + "examples/visual_gen/test_visual_gen.py::test_flux1_lpips_against_golden": 61.382666666666665, + "examples/visual_gen/test_visual_gen.py::test_flux2_example": 370.0974792746114, + "examples/visual_gen/test_visual_gen.py::test_flux2_lpips_against_golden": 85.6677142857143, + "examples/visual_gen/test_visual_gen.py::test_flux2_reference_image_example": 289.24313513513516, + "examples/visual_gen/test_visual_gen.py::test_ltx2_cuda_graph_lpips_matches_eager": 293.57128571428575, + "examples/visual_gen/test_visual_gen.py::test_ltx2_example": 382.7444775725594, + "examples/visual_gen/test_visual_gen.py::test_ltx2_lpips_against_golden": 189.37857142857143, + "examples/visual_gen/test_visual_gen.py::test_qwen_image_example": 224.7498372093023, + "examples/visual_gen/test_visual_gen.py::test_qwenimage_cuda_graph_lpips_against_golden": 79.9952, + "examples/visual_gen/test_visual_gen.py::test_qwenimage_lpips_against_golden": 86.78375, + "examples/visual_gen/test_visual_gen.py::test_visual_gen_api_walkthrough": 239.43309943181816, + "examples/visual_gen/test_visual_gen.py::test_visual_gen_quickstart": 198.98814145658264, + "examples/visual_gen/test_visual_gen.py::test_wan21_t2v_lpips_against_golden": 36.528571428571425, + "examples/visual_gen/test_visual_gen.py::test_wan22_t2v_lpips_against_golden": 63.1655, + "examples/visual_gen/test_visual_gen.py::test_wan_t2v_example": 571.5687341772151, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[attn2d_2x2]": 953.7801428571429, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[cfg2_ulysses2]": 270.7155, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_multi_gpu[ulysses4]": 1041.747, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[cfg2_tp2]": 151.057, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2]": 73.04014285714285, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_attn2d_2x1]": 275.8075, + "examples/visual_gen/test_visual_gen_multi_gpu.py::test_wan22_t2v_lpips_against_golden_tp[tp2_ulysses2]": 181.01357142857142, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_changed_final_token_reuse_cuda_graph[v1]": 20.15875, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_changed_final_token_reuse_cuda_graph[v2]": 29.40618181818182, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_context_logits_after_final_token_reuse[v1]": 11.5116, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_context_logits_after_final_token_reuse[v2]": 16.64923076923077, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph[v1]": 9.9382, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph[v2]": 15.367846153846154, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph_tp2[v1]": 32.907, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_final_token_reuse_cuda_graph_tp2[v2]": 40.815, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_guided_decoding_after_final_token_reuse[v1]": 10.742833333333333, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_guided_decoding_after_final_token_reuse[v2]": 15.2503125, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_zero_runtime_draft_speculation_after_final_token_reuse[v1]": 21.316636363636363, + "kv_cache/test_final_single_token_context_cuda_graph.py::test_zero_runtime_draft_speculation_after_final_token_reuse[v2]": 26.2569375, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_batch_generation": 18.130119363395224, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_cold_start": 21.948513368983956, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_field_completeness": 12.267211796246649, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_full_block_reuse": 19.860912, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_long_context": 29.62158510638298, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_partial_block_reuse": 25.294635135135138, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_rapid_fire": 13.005842391304348, + "kv_cache/test_kv_cache_iteration_stats.py::TestKvCacheIterationStats::test_shared_prefix": 19.404803278688522, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_chunked_draft_tokens": 291.17222857142855, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_draft_tokens": 126.47582857142858, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2DSv3Lite::test_mtp_eviction": 31.50208653846154, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_batch_size_limited": 130.98199723756906, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_block_reuse": 162.45401400560226, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill": 181.319, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_eviction_block_reuse": 20.081730245231608, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_multi_request": 157.2019005524862, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_chunked_prefill_with_eviction": 20.046337950138504, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[cuda_graph]": 17.19818487394958, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction[no_cuda_graph]": 28.09476404494382, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_overlap": 17.407029810298102, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_eviction_with_block_reuse": 43.698, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[non_overlap]": 135.2105218579235, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_overlap_scheduler[overlap]": 154.92498382749326, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_partial_block_reuse": 170.93294623655913, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_token_budget_limited": 133.1101049723757, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2Llama::test_v2_vs_v1_basic": 135.72120055710306, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_chunked_prefill": 133.00545658263306, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_eviction": 46.97228690807799, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_multi_adapter_v2": 164.67938292011019, + "kv_cache/test_kv_cache_v2_scheduler.py::TestKVCacheV2LoRA::test_lora_v2": 190.69342222222224, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[guaranteed-chunked]": 159.4352530120482, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[max-util-chunked]": 159.47862611275963, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[no-overlap-chunked]": 169.5043982300885, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-chunked]": 162.48168338557994, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[offload-no-chunked]": 162.01903592814372, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[python-scheduler]": 168.00891788856305, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix[swa-chunked]": 133.72491875, + "kv_cache/test_prefix_aware_scheduling.py::TestServePrefixAwareScheduling::test_multi_round_qa_shared_prefix_smoke": 132.2567359050445, + "llmapi/test_llm_api_connector.py::test_connector_async_onboard[False]": 17.844191326530613, + "llmapi/test_llm_api_connector.py::test_connector_async_onboard[True]": 17.94226785714286, + "llmapi/test_llm_api_connector.py::test_connector_async_save[False]": 17.79670512820513, + "llmapi/test_llm_api_connector.py::test_connector_async_save[True]": 12.164752551020408, + "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[False]": 23.325279792746116, + "llmapi/test_llm_api_connector.py::test_connector_disagg_prefill[True]": 23.173549872122763, + "llmapi/test_llm_api_connector.py::test_connector_e2e_persistent_cache": 21.657964194373402, + "llmapi/test_llm_api_connector.py::test_connector_multi_request": 9.827329048843188, + "llmapi/test_llm_api_connector.py::test_connector_priorities": 11.924429667519181, + "llmapi/test_llm_api_connector.py::test_connector_priorities_default": 11.87047193877551, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[attention_dp]": 3.934010101010101, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[beam_search]": 9.53984693877551, + "llmapi/test_llm_api_connector.py::test_connector_rejects_unsupported_config[host_offloading]": 11.150828205128207, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[False]": 14.05459383033419, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output[True]": 13.803569974554708, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[False]": 11.044869897959185, + "llmapi/test_llm_api_connector.py::test_connector_scheduler_output_chunked_context[True]": 10.860375, + "llmapi/test_llm_api_connector.py::test_connector_simple[False]": 11.83360358056266, + "llmapi/test_llm_api_connector.py::test_connector_simple[True]": 11.803266666666666, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-bart-large-cnn]": 31.893891414141414, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-bart-large-cnn]": 15.499701803051318, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-bart-large-cnn]": 22.250857142857143, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 28.036251908396945, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-bart-large-cnn]": 6.71089847715736, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-bart-large-cnn]": 13.861859174964438, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-bart-large-cnn]": 22.041833333333333, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-greedy-bart-large-cnn]": 19.904857142857143, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 7.540977157360405, + "llmapi/test_llm_api_pytorch_bart.py::test_bart_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-bart-large-cnn]": 15.9137, + "llmapi/test_llm_api_pytorch_bart.py::test_mbart_pytorch_generate_encoder_decoder_end_to_end": 18.597017811704834, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[cudagraph]": 83.4373853211009, + "llmapi/test_llm_api_pytorch_moe_lora.py::test_qwen_moe_routed_expert_multi_lora_varying_ranks[eager]": 81.5805, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-off-greedy-t5-small]": 5.8821772151898735, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 11.94230964467005, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-flan-t5-xl]": 36.057, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-overlap-t5-small]": 18.94985714285714, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-base]": 19.276714285714288, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-beam2-t5-small]": 12.308420903954802, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-overlap-t5-small]": 11.595129760225669, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-t5-small]": 5.213745547073791, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v1-cuda-graph-on-greedy-tp2-t5-small]": 21.989346534653464, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-off-greedy-byt5-small]": 18.854222222222223, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.96542857142857, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-overlap-t5-small]": 18.41714285714286, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[bf16-kv-v2-cuda-graph-on-greedy-t5-small]": 5.065276649746193, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 18.676571428571428, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v1-cuda-graph-on-beam2-t5-small]": 18.371142857142857, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.455428571428573, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp16-kv-v2-cuda-graph-on-greedy-t5-small]": 17.964, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-flan-t5-small]": 17.551571428571428, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v1-cuda-graph-on-beam2-t5-small]": 18.958428571428573, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-flan-t5-small]": 18.94242857142857, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_end_to_end[fp32-kv-v2-cuda-graph-on-greedy-t5-small]": 18.021428571428572, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_context_generation_batch": 5.631847715736041, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-flan-t5-small]": 15.022181818181817, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-beam2-batch2-t5-small]": 15.552, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v1-decoder-cuda-graph-on-greedy-batch2-t5-small]": 5.275149367088607, + "llmapi/test_llm_api_pytorch_t5.py::test_t5_pytorch_generate_encoder_decoder_mixed_encoder_lengths_batch[bf16-kv-v2-decoder-cuda-graph-on-greedy-batch2-t5-small]": 16.256125, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_beam_search[bf16-kv-v1-decoder-graphs-on-beam2]": 9.641889502762432, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_beam_search[fp32-kv-v1-graphs-off-beam2]": 8.5922, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[bf16-kv-v1-decoder-graphs-on-greedy]": 12.066636363636364, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[bf16-kv-v2-decoder-graphs-on-greedy]": 19.83623324396783, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp16-kv-v1-graphs-off-greedy]": 10.803600000000001, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v1-graphs-off-greedy-tp2]": 25.9721875, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v1-graphs-requested-greedy]": 5.953636363636364, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_feature_combinations[fp32-kv-v2-graphs-off-greedy]": 6.5942, + "llmapi/test_llm_api_pytorch_whisper.py::test_whisper_pytorch_transcribe_end_to_end": 14.981148247978437, + "llmapi/test_llm_examples.py::test_llmapi_chat_example": 67.295, + "llmapi/test_llm_examples.py::test_llmapi_example_distributed_tp2": 53.464285714285715, + "llmapi/test_llm_examples.py::test_llmapi_example_guided_decoding": 59.278842083675144, + "llmapi/test_llm_examples.py::test_llmapi_example_inference": 66.28026499589154, + "llmapi/test_llm_examples.py::test_llmapi_example_inference_async": 46.420170701682395, + "llmapi/test_llm_examples.py::test_llmapi_example_inference_async_streaming": 52.58203203285421, + "llmapi/test_llm_examples.py::test_llmapi_example_logits_processor": 45.783905916187344, + "llmapi/test_llm_examples.py::test_llmapi_example_multilora": 59.31795570139459, + "llmapi/test_llm_examples.py::test_llmapi_kv_cache_connector[Qwen2-0.5B]": 88.863, + "llmapi/test_llm_examples.py::test_llmapi_quickstart": 45.94649178307313, + "llmapi/test_llm_examples.py::test_llmapi_quickstart_atexit": 277.5743333333333, + "llmapi/test_llm_examples.py::test_llmapi_server_example": 155.23228571428572, + "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_mtp": 218.2092435897436, + "llmapi/test_llm_examples.py::test_llmapi_speculative_decoding_ngram": 92.54680097087378, + "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_context": 0.9749723756906077, + "perf/host_perf/test_module_resource_manager.py::test_kv_cache_prepare_generation": 0.919544474393531, + "perf/host_perf/test_module_sampler.py::test_sampler_update_greedy[greedy_bs8]": 1.1856621253405994, + "perf/host_perf/test_module_sampler.py::test_sampler_update_stop_words[stopwords_bs32]": 1.2631570247933885, + "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_gen_only_bs8]": 0.8479316939890711, + "perf/host_perf/test_module_scheduler.py::test_scheduler_production[production_mixed_32gen_4ctx]": 0.9042729729729729, + "perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:8192]": 243.04695381062356, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 1999.384, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 408.2405, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 960.0195714285715, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 379.282625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2356.765423255814, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 456.45875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 2758.6535, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 387.9955, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 2762.70475, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con4_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 198.09625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 572.877202764977, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1376.504375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2174.338375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 512.32025, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2131.6131612903223, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 316.531625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 407.063537037037, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 4896.811888888889, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2131.481125, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1495.524375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 2059.856375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1268.812375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 2882.679875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb416_mtp3_ccb-NIXL]": 1273.481375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1972.6055, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-ctx_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 466.771375, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 1304.3386, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_2_nodes_grace_blackwell-r1_fp4_v2_tep8_mtp3]": 541.446, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_dep8_mtp1_8k1k]": 651.103, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 398.36628571428577, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_dep4_mtp1_8k1k]": 905.3361290322581, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_1k8k]": 950.800625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tep4_mtp3_8k1k]": 561.985, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_1k8k]": 601.254875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp4_v2_grace_blackwell-r1_fp4_v2_tp4_mtp3_8k1k]": 402.47142396313365, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_ad_blackwell-r1_fp8_ad_ws8_1k1k]": 715.4331428571428, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_r1_fp8_blackwell-r1_fp8_tp8_6k1k]": 1442.0042857142857, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_dep8_mtp1_8k1k]": 858.155875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_blackwell-v32_fp4_tep8_mtp3_8k1k]": 380.380875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_dep4_mtp1_8k1k]": 917.161625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-deepseek_v32_fp4_grace_blackwell-v32_fp4_tep4_mtp3_8k1k]": 489.380875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_deepseek_v32_fp4_2_nodes_grace_blackwell-dsv32_fp4_dep8_trtllm_lpc_mnnvl_8k1k]": 453.6847142857143, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_adp_2k1k]": 590.84, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_235b_a22b_fp8_hopper-qwen3_235b_a22b_fp8_tp4_ep4_cutlass_8k1k]": 946.8511428571428, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-dynamo_qwen3_32b_fp8_hopper-qwen3_32b_fp8_tp2_6k1k]": 768.5237142857143, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gemma4_26b_a4b_nvfp4_blackwell-gemma4_26b_a4b_nvfp4_tp1_1k1k]": 322.5816615384615, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 578.0965, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_2_nodes_grace_blackwell-glm5_fp4_tep8_mtp3_8k1k]": 574.7941428571428, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-glm5_fp4_blackwell-glm5_fp4_dep8_mtp1_8k1k]": 688.625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 595.070625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_dep4_1k8k]": 1280.064875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tep2_1k8k]": 802.94175, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp1_mtp0_8k1k]": 589.688, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-gpt_oss_120b_fp4_grace_blackwell-gpt_oss_fp4_tp2_1k8k]": 501.079625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_deepseek_v3_lite-v3lite_fp8_bs8_128_256]": 569.2608436657681, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b-llama8b_fp16_bs8_128_256]": 264.4596918767507, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-host_perf_llama8b_spec_decode-llama8b_spec_bs1_128_128]": 177.14550555555553, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_2_nodes_grace_blackwell-k25_thinking_fp4_dep8_32k8k]": 1465.1101666666668, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_32k8k]": 1438.051875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_dep8_8k1k]": 764.6005, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_32k8k]": 562.570625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_blackwell-k25_thinking_fp4_tep8_8k1k]": 340.288625, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_dep4_8k1k]": 837.4493411214953, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-k25_thinking_fp4_grace_blackwell-k25_thinking_fp4_tep4_8k1k]": 447.6755, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_1000_1000]": 462.81728571428573, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-llama_v3_3_70b_instruct_fp4_blackwell-llama70b_fp4_tp4_512_32]": 250.122875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_8k1k]": 708.9645, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_dep8_mtp3_8k1k]": 599.9765, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tep4_mtp3_8k1k]": 567.198875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-qwen3_5_397b_fp4_blackwell-qwen3_5_397b_fp4_tp4_8k1k]": 518.17875, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_ad_blackwell-super_ad_ws4_1k1k]": 579.10435, + "perf/test_perf_sanity.py::test_e2e[aggr_upload-super_mtp_ad_blackwell-super_mtp_ad_ws4_1k1k]": 541.9476666666666, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 4236.98, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2463.5651428571427, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_32k4k_con2048_ctx1_dep4_gen1_dep32_eplb288_mtp1_ccb-NIXL]": 3260.8825714285713, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_deepseek-v32-fp4_8k1k_con4096_ctx1_dep4_gen1_dep32_eplb256_mtp0_ccb-NIXL]": 3234.713142857143, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 865.6539590909091, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 2143.3998571428574, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 2017.474, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 6895.875, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2307.246111111111, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1396.0565, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 2622.707125, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1470.1127142857142, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1339.9375714285713, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 3237.387563218391, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-e2e-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 1942.96, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con1536_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 871.0367857142857, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 508.911, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-b200_deepseek-r1-fp4_8k1k_con256_ctx1_dep4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 554.807, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con128_ctx1_pp8_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 2544.6345714285712, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_128k8k_con1_ctx1_pp8_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 862.036375, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con1024_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 539.3786, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 427.9478, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-r1-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp1_ccb-NIXL]": 885.85225, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 635.96075, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep4_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 871.6156, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_32k4k_con256_ctx1_dep8_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 579.8865, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_deepseek-v32-fp4_8k1k_con1_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 1007.6004, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con1024_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 322.9055, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con128_ctx1_tp1_gen1_tp4_eplb0_mtp0_ccb-NIXL]": 298.158, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_gpt-oss-120b-fp4_8k1k_con512_ctx1_tp1_gen1_dep2_eplb0_mtp0_ccb-NIXL]": 279.817375, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 865.159, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1024_ctx1_tp1_gen1_dep8_eplb0_mtp0_ccb-NIXL]": 497.62114285714284, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con1_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 313.13625, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb200_qwen3-235b-fp4_8k1k_con64_ctx1_tp1_gen1_tep4_eplb0_mtp0_ccb-NIXL]": 331.31875, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-r1-fp4_128k8k_con256_ctx1_pp4_gen1_dep8_eplb0_mtp1_ccb-NIXL]": 2963.189, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con180_ctx3_dep4_gen1_dep32_eplb384_mtp3_ccb-NIXL]": 1329.4505, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con4301_ctx12_dep4_gen1_dep8_eplb384_mtp1_ccb-NIXL]": 1577.9463333333333, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con666_ctx6_dep4_gen1_dep16_eplb384_mtp3_ccb-NIXL]": 1356.0965, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_deepseek-v4-pro-fp4_8k1k_con8_ctx1_dep4_gen4_tep8_eplb0_mtp3_ccb-NIXL]": 1260.772, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1024_ctx1_dep2_gen1_dep8_eplb256_mtp1_ccb-NIXL]": 1073.543, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con1_ctx1_dep2_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 605.6804285714285, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_glm-5-fp4_8k1k_con512_ctx1_dep2_gen1_dep32_eplb0_mtp3_ccb-NIXL]": 598.488, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4096_ctx1_dep4_gen1_dep16_eplb0_mtp0_ccb-NIXL]": 878.816, + "perf/test_perf_sanity.py::test_e2e[disagg_upload-gen_only-gb300_kimi-k25-thinking-fp4_8k1k_con4_ctx1_dep4_gen1_tep8_eplb0_mtp3_ccb-NIXL]": 469.214023923445, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-flux2_blackwell-flux2_fp8_cfg1_ulysses4_teacache_on]": 333.64025, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_i2v_cfg2_ulysses4_compile_on]": 536.826875, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_2stage_bf16_t2v_cfg2_ulysses4_compile_on]": 476.725875, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-ltx2_blackwell-ltx2_nvfp4_i2v_cfg2_ulysses4_compile_on]": 472.473875, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan21_t2v_14b_blackwell-wan21_14b_nvfp4_trtllm_cfg2_ulysses4_teacache_on]": 511.6845, + "perf/test_visual_gen_perf_sanity.py::test_visual_gen_e2e[vg_upload-wan22_i2v_a14b_blackwell-wan22_i2v_a14b_nvfp4_trtllm_cfg2_ulysses4]": 458.00059999999996, + "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp1_4instances]": 103.69261616161616, + "ray_orchestrator/RL/test_rl_perf_reproduce.py::test_rl_perf_reproduce[tp2_2instances]": 107.46637373737374, + "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-GUARANTEED_NO_EVICT-pytorch-stress-test]": 820.9262857142857, + "stress_test/stress_test.py::test_run_stress_test[llama-v3-8b-instruct-hf_tp1-stress_time_300s_timeout_450s-MAX_UTILIZATION-pytorch-stress-test]": 736.4245714285714, + "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 63.3935, + "test_e2e.py::test_draft_token_tree_quickstart_advanced_eagle3_depth_1_tree[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 59.22325, + "test_e2e.py::test_get_ci_container_port": 1.294469543147208, + "test_e2e.py::test_openai_chat_example[pytorch]": 568.4271944444445, + "test_e2e.py::test_openai_chat_guided_decoding[meta-llama/Llama-3.1-8B-Instruct]": 72.82539402985076, + "test_e2e.py::test_openai_chat_guided_decoding[openai/gpt-oss-120b]": 170.5241832884097, + "test_e2e.py::test_openai_chat_harmony": 180.47430188679243, + "test_e2e.py::test_openai_chat_harmony_perf_metrics": 168.00521202531647, + "test_e2e.py::test_openai_chat_multimodal_example": 128.08525063938617, + "test_e2e.py::test_openai_chat_with_logit_bias[torch_sampler]": 85.82997073170732, + "test_e2e.py::test_openai_chat_with_logit_bias[trtllm_sampler]": 85.85354501216544, + "test_e2e.py::test_openai_completions_example[pytorch]": 255.78602307692307, + "test_e2e.py::test_openai_completions_with_logit_bias[torch_sampler]": 83.29217518248174, + "test_e2e.py::test_openai_completions_with_logit_bias[trtllm_sampler]": 83.41751707317073, + "test_e2e.py::test_openai_health": 69.34942857142856, + "test_e2e.py::test_openai_kv_cache_contamination": 1558.114663265306, + "test_e2e.py::test_openai_lora": 184.53052987012987, + "test_e2e.py::test_openai_misc_example[pytorch]": 338.3785968992248, + "test_e2e.py::test_openai_mmencoder_example": 132.4975871794872, + "test_e2e.py::test_openai_perf_metrics": 83.57704303797469, + "test_e2e.py::test_openai_post_processor": 367.14763144329896, + "test_e2e.py::test_openai_prometheus": 91.01292583120204, + "test_e2e.py::test_openai_reasoning[pytorch]": 253.63865025906736, + "test_e2e.py::test_openai_responses": 318.4291543026706, + "test_e2e.py::test_openai_responses_entrypoint": 180.99946272493574, + "test_e2e.py::test_openai_tool_call": 92.28577581863979, + "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-120B-gpt_oss/gpt-oss-120b]": 405.7005, + "test_e2e.py::test_ptp_quickstart_advanced[GPT-OSS-20B-gpt_oss/gpt-oss-20b]": 74.43636476426799, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-BF16-llama-3.1-model/Meta-Llama-3.1-8B]": 78.70912158808933, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-FP8-llama-3.1-model/Llama-3.1-8B-Instruct-FP8]": 60.669947300771206, + "test_e2e.py::test_ptp_quickstart_advanced[Llama3.1-8B-NVFP4-nvfp4-quantized/Meta-Llama-3.1-8B]": 80.62433642384106, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B-Qwen3/Qwen3-30B-A3B]": 156.90823631840797, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_fp8_hf-Qwen3/saved_models_Qwen3-30B-A3B_fp8_hf]": 116.35483084577113, + "test_e2e.py::test_ptp_quickstart_advanced[Qwen3-30B-A3B_nvfp4_hf-Qwen3/saved_models_Qwen3-30B-A3B_nvfp4_hf]": 94.89845024875622, + "test_e2e.py::test_ptp_quickstart_advanced_bs1": 97.3342, + "test_e2e.py::test_ptp_quickstart_advanced_deepseek_v3_lite_4gpus_adp_balance[DeepSeek-V3-Lite-FP8-DeepSeek-V3-Lite/fp8]": 95.85601020408163, + "test_e2e.py::test_ptp_quickstart_advanced_eagle3[Llama-3.1-8b-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct-EAGLE3-LLaMA3.1-Instruct-8B]": 75.93503878116343, + "test_e2e.py::test_ptp_quickstart_advanced_mixed_precision": 90.59626630434782, + "test_e2e.py::test_ptp_quickstart_advanced_mtp[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 168.06148467966574, + "test_e2e.py::test_ptp_quickstart_advanced_mtp_eagle[DeepSeek-V3-Lite-BF16-DeepSeek-V3-Lite/bf16]": 250.64657452574528, + "test_e2e.py::test_ptp_quickstart_advanced_ngram[Llama-3.1-8B-Instruct-llama-3.1-model/Llama-3.1-8B-Instruct]": 70.2027741046832, + "test_e2e.py::test_ptp_quickstart_bert[TRTLLM-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 20.47236647955092, + "test_e2e.py::test_ptp_quickstart_bert[VANILLA-BertForSequenceClassification-bert/bert-base-uncased-yelp-polarity]": 6.805838813151564, + "test_e2e.py::test_trtllm_bench_help_sanity[meta-llama/Llama-3.1-8B]": 46.12830721003135, + "test_e2e.py::test_trtllm_bench_invalid_token_pytorch[TinyLlama-1.1B-Chat-v1.0-TinyLlama-1.1B-Chat-v1.0]": 77.25574680306904, + "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-non-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 89.06281155015198, + "test_e2e.py::test_trtllm_bench_iteration_log[PyTorch-streaming-meta-llama/Llama-3.1-8B-llama-3.1-model/Meta-Llama-3.1-8B]": 88.87241614906831, + "test_e2e.py::test_trtllm_bench_llmapi_launch[pytorch_backend-llama-v3-llama3-8b]": 123.27977142857144, + "test_e2e.py::test_trtllm_bench_mgmn": 113.6422857142857, + "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-False-False]": 117.62252560386473, + "test_e2e.py::test_trtllm_bench_pytorch_backend_sanity[meta-llama/Llama-3.1-8B-llama-3.1-8b-instruct-hf-fp8-True-True]": 67.63684076433121, + "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-]": 115.25042385786803, + "test_e2e.py::test_trtllm_bench_request_rate_and_concurrency[enable_concurrency-enable_request_rate]": 32.17077581120944, + "test_e2e.py::test_trtllm_serve_example": 135.66885714285712, + "test_e2e.py::test_trtllm_serve_lora_example": 93.26694545454545, + "test_e2e.py::test_trtllm_serve_multimodal_example": 176.18389285714287, + "test_e2e.py::test_trtllm_serve_top_logprobs[pytorch]": 85.50150381679389, + "test_fmha.py::test_fmha": 2383.811928571429, + "thirdparty/test_cmake_third_party.py::test_cmake_listfiles": 1.309677749360614, + "thirdparty/test_git_modules.py::test_gitmodules": 1.2179113924050633, + "unittest/_torch/attention": 1427.1057219730942, + "unittest/_torch/attention --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py --ignore=unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1447.3447479452057, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_kernel.py": 26.8552816091954, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_module.py": 132.06273595505618, + "unittest/_torch/attention/sparse/deepseek_v4/test_compressor_tf32.py": 23.715, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_cache_manager.py": 54.658080332409966, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_indices_transform.py": 58.47613623978202, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_o_proj.py": 37.03518105849582, + "unittest/_torch/attention/sparse/deepseek_v4/test_deepseek_v4_sparse_mla.py": 55.00545303867403, + "unittest/_torch/attention/sparse/dsa/test_dsa_fp4_indexer.py": 23.768575757575757, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_cute_dsl_fp8_indexer_q_gemm_rope_fp4_matches_unfused": 47.0700824742268, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_dsv32_prefill_shape": 24.42627, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_matches_deepgemm": 20.244656565656566, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_noncontiguous_split": 2.1675763440860214, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_fused_cat_fp4_shape_dispatch": 22.984191919191918, + "unittest/_torch/attention/sparse/test_cpp_custom_ops.py::test_indexer_k_cache_gather_contiguous_fp4": 20.064323232323233, + "unittest/_torch/attention/sparse/test_cute_dsl_fp4_paged_mqa_logits.py": 725.6988395721925, + "unittest/_torch/attention/sparse/test_cute_dsl_fp8_paged_mqa_logits.py": 344.4336289473684, + "unittest/_torch/attention/sparse/test_cute_dsl_gvr_topk_decode.py": 1976.4200219178083, + "unittest/_torch/attention/sparse/test_sparse_mla_forward.py": 122.47299180327869, + "unittest/_torch/attention/test_attention_mla.py": 133.47027261904762, + "unittest/_torch/auto_deploy": 25.987419558359623, + "unittest/_torch/compilation": 26.45571775417299, + "unittest/_torch/custom_ops/test_deepseek_v4_q_norm.py": 23.911654494382024, + "unittest/_torch/debugger": 21.257522302158275, + "unittest/_torch/distributed": 20.148352380952378, + "unittest/_torch/executor": 312.95409486780716, + "unittest/_torch/executor/test_async_transfer_manager.py": 26.81270284237726, + "unittest/_torch/executor/test_disagg_index_mapper_early_release.py": 27.300713917525773, + "unittest/_torch/executor/test_error_classification.py": 25.742147582697203, + "unittest/_torch/executor/test_hang_detector_kill.py": 41.70411578947368, + "unittest/_torch/executor/test_kv_cache_budget_split.py": 26.940844387755103, + "unittest/_torch/executor/test_kv_cache_compression_manager.py": 27.190636597938145, + "unittest/_torch/executor/test_kv_cache_estimation.py": 27.18597461928934, + "unittest/_torch/executor/test_kv_cache_v2_capacity_only.py": 25.351969072164948, + "unittest/_torch/executor/test_kv_pool_rebalance.py": 27.080116279069767, + "unittest/_torch/executor/test_model_loader_mx.py": 23.324768304914745, + "unittest/_torch/executor/test_overlap_scheduler.py": 762.4326900584796, + "unittest/_torch/executor/test_scheduler_serializable_output.py": 25.51113810741688, + "unittest/_torch/flashinfer/test_trtllm_flashinfer_symbol_collision.py": 1010.8754930362117, + "unittest/_torch/lora": 21.032186781609195, + "unittest/_torch/memory": 20.313926751592355, + "unittest/_torch/misc": 207.61796048632218, + "unittest/_torch/misc/test_autotuner.py::test_autotuner_distributed_strategy": 42.2076, + "unittest/_torch/modeling -k \"modeling_gemma3\"": 246.71202580645163, + "unittest/_torch/modeling -k \"modeling_gpt_oss\"": 98.51157242339833, + "unittest/_torch/modeling -k \"modeling_llama\"": 134.55647529069768, + "unittest/_torch/modeling -k \"modeling_mixtral\"": 77.18025104022192, + "unittest/_torch/modeling -k \"modeling_nemotron_nano_v2_vl\"": 457.9309539641944, + "unittest/_torch/modeling -k \"modeling_nemotron_nas\"": 39.4559902676399, + "unittest/_torch/modeling -k \"modeling_out_of_tree\"": 143.6954725014784, + "unittest/_torch/modeling -k \"modeling_phi3\"": 33.97074209245742, + "unittest/_torch/modeling -k \"modeling_qwen\"": 701.1010346820809, + "unittest/_torch/modeling -k \"modeling_qwen_moe\"": 298.2582273838631, + "unittest/_torch/modeling -k \"modeling_siglip\"": 190.55107537688443, + "unittest/_torch/modeling -k \"modeling_speculative\"": 27.12980684596577, + "unittest/_torch/modeling -k \"modeling_vila\"": 42.61871392405063, + "unittest/_torch/modeling -k \"modeling_whisper\"": 22.399005917159766, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_multimodal_26b_dummy": 174.59721359223303, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_26b_dummy": 176.25729368575625, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_31b_dummy": 207.95196536796536, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy": 178.88090070921987, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[256]": 182.0266661991585, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e2b_dummy_small_max_seq_len[512]": 207.8059829302987, + "unittest/_torch/modeling/test_gemma4_e2e_dummy.py::test_e2e_text_e4b_dummy": 188.1576099585062, + "unittest/_torch/modeling/test_gemma4_multimodal.py": 26.365072625698325, + "unittest/_torch/modeling/test_modeling_afmoe.py": 70.01261643835618, + "unittest/_torch/modeling/test_modeling_cohere2.py": 77.30710997442456, + "unittest/_torch/modeling/test_modeling_cosmos3.py::TestCosmos3::test_all": 25.3536120906801, + "unittest/_torch/modeling/test_modeling_deepseekv4.py": 57.31769696969697, + "unittest/_torch/modeling/test_modeling_exaone_moe.py": 252.087638483965, + "unittest/_torch/modeling/test_modeling_gemma4.py": 80.82831623931624, + "unittest/_torch/modeling/test_modeling_gemma4_unified.py": 20.30774293785311, + "unittest/_torch/modeling/test_modeling_llava_next.py::TestLlavaNext::test_all": 35.675928030303034, + "unittest/_torch/modeling/test_modeling_llava_next.py::test_llava_next_expand_prompt_token_ids_for_mm": 14.221383165829145, + "unittest/_torch/modeling/test_modeling_minicpmv4_6.py": 28.253985454545454, + "unittest/_torch/modeling/test_modeling_mistral.py": 40.019712820512815, + "unittest/_torch/modeling/test_modeling_nemotron_h.py::test_nemotron_h_sanity": 181.17747040498443, + "unittest/_torch/modeling/test_modeling_parakeet.py": 43.805466836734695, + "unittest/_torch/modeling/test_modeling_pixtral.py": 35.962048969072164, + "unittest/_torch/modeling/test_modeling_pixtral.py::test_tensor_parallelism": 42.47271568627451, + "unittest/_torch/modeling/test_modeling_qwen2_5vl.py::TestQwen2_5_VL::test_all": 76.01826083188908, + "unittest/_torch/modeling/test_modeling_qwen3_5_vl.py::TestQwen3_5VL::test_all": 71.38927666666667, + "unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py::TestQwen3_5MoeVL::test_all": 79.15871782945737, + "unittest/_torch/modeling/test_modeling_qwen3vl.py::TestQwen3VL::test_all": 96.50474396551724, + "unittest/_torch/modeling/test_modeling_qwen3vl.py::test_qwen3vl_init_preserves_caller_quant_config": 11.537473136915077, + "unittest/_torch/modeling/test_modeling_qwen3vl_moe.py::TestQwen3VLMoe::test_all": 82.38036465517241, + "unittest/_torch/modeling/test_modeling_radio.py": 29.20512853470437, + "unittest/_torch/modeling/test_modeling_step3p7.py": 28.51029411764706, + "unittest/_torch/modeling/test_modeling_step3p7vl.py": 40.39187341772151, + "unittest/_torch/modeling/test_multimodal_encoder_graph.py": 20.74985975609756, + "unittest/_torch/modeling/test_multimodal_encoder_mixin.py": 25.43182005141388, + "unittest/_torch/modeling/test_nemotron_nano_preprocessing.py": 63.15041176470588, + "unittest/_torch/models/checkpoints": 63.97590512820513, + "unittest/_torch/models/test_qwen3_next_moe_quant.py": 25.30277272727273, + "unittest/_torch/modules/dwdp/test_dwdp_fixup_moe_backends.py": 25.54453746770026, + "unittest/_torch/modules/dwdp/test_dwdp_manager.py": 25.44528165374677, + "unittest/_torch/modules/dwdp/test_dwdp_mapping.py": 25.49276546391753, + "unittest/_torch/modules/dwdp/test_dwdp_peer_ranges.py": 25.482807291666667, + "unittest/_torch/modules/fused_moe/test_deepgemm_fused_expand_quant.py": 22.72942074927954, + "unittest/_torch/modules/fused_moe/test_deepgemm_fused_gather_finalize.py": 23.010556603773583, + "unittest/_torch/modules/fused_ops/test_fused_qk_norm_rope_gate.py": 24.518203296703298, + "unittest/_torch/modules/fused_ops/test_gelu_tanh_mul_fp4_quant.py": 25.7476404494382, + "unittest/_torch/modules/fused_ops/test_rmsnorm_fp4_quant.py": 24.488736263736264, + "unittest/_torch/modules/fused_ops/test_rmsnorm_residual_add.py": 24.566434065934068, + "unittest/_torch/modules/mamba": 743.5672699507388, + "unittest/_torch/modules/moe/test_cute_dsl_b12x_moe_backend.py": 21.25159801488834, + "unittest/_torch/modules/moe/test_megamoe_streaming_load.py": 21.72980769230769, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 95.0835, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTEDSL\"": 261.44885714285715, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"CUTLASS\"": 1145.8342608695652, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 45.4305, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DEEPGEMM\"": 62.880714285714284, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"DENSEGEMM\"": 664.1701428571429, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_CUTEDSL\"": 121.23383333333332, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"MEGAMOE_DEEPGEMM\"": 28.824571428571428, + "unittest/_torch/modules/moe/test_moe_backend.py::test_moe_backend -k \"TRTLLM\"": 121.97892307692308, + "unittest/_torch/modules/moe/test_moe_backend.py::test_trtllm_bf16_unquantized_moe": 934.3070607344632, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm": 260.6249773755656, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_boundary": 227.2692, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_moe_comm_postquant": 123.30966502463053, + "unittest/_torch/modules/moe/test_moe_comm.py::TestMoEComm::test_nccl_ep_cuda_graph_replay_uses_updated_routing": 45.36088387096775, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTEDSL\"": 96.90683838383839, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES and not W4A8 and not MXFP8\"": 79.44716666666667, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8 and not FP8_BLOCK_SCALES\"": 71.319, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and FP8_BLOCK_SCALES\"": 60.91212, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and MXFP8 and not W4A8\"": 74.81355555555557, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and NVFP4\"": 72.06288, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A16_MXFP4\"": 70.43882653061225, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_AWQ\"": 62.54622, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_FP8\"": 80.16814141414142, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W4A8_MXFP4_MXFP8\"": 107.48528282828282, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"CUTLASS and W8A16\"": 78.97271573604061, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 79.17475757575757, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MARLIN\"": 81.26303157894736, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_CUTEDSL\"": 62.499848484848485, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"MEGAMOE_DEEPGEMM\"": 54.97358585858586, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and FP8_BLOCK_SCALES\"": 61.35357575757576, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and NVFP4 and not W4A8\"": 58.35683838383838, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A16_MXFP4\"": 66.28399, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_MXFP4_MXFP8\"": 52.592858585858586, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu -k \"TRTLLM and W4A8_NVFP4_FP8\"": 151.46551515151515, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 63.586188118811876, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 47.33725742574257, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 56.93824752475248, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 80.86554455445544, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=DEEPEP-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.643316831683165, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_CUTEDSL-quant=NVFP4-routing=DeepSeekV3]": 58.67727722772277, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=IGNORE-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=MEGAMOE_DEEPGEMM-quant=W4A8_MXFP4_MXFP8-routing=DeepSeekV3]": 55.511158415841585, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 26.81741584158416, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 26.794633663366337, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e256_k6_h4096_i2048-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 26.845336633663365, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 48.804900990099014, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e60_k4_h2048_i1408-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.747702970297034, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 46.98811881188119, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb": 48.077839999999995, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_CUTEDSL\"": 57.46130693069307, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"MEGAMOE_DEEPGEMM\"": 52.91237623762377, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb -k \"not None\"": 25.53242, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 47.52238613861386, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 48.06145544554455, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_multi_gpu_eplb[parallel=DEP-comm=NVLINK_ONE_SIDED-e8_k2_h512_i512-slots=16-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 48.202405940594055, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTEDSL and not MEGAMOE_CUTEDSL\"": 89.27855492957747, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and None\"": 27.112461538461538, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"CUTLASS and not None\"": 149.8221215805471, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DEEPGEMM and not MEGAMOE_DEEPGEMM\"": 50.51278611111111, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"DENSEGEMM\"": 454.09614, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_CUTEDSL\"": 114.45545983379502, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu -k \"MEGAMOE_DEEPGEMM\"": 42.27300540540541, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTEDSL-quant=NVFP4-routing=Renormalize]": 65.58272594752187, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=FP8-routing=Renormalize]": 32.970935754189945, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 31.079560344827588, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W4A8_MXFP4_MXFP8-routing=Renormalize]": 29.457871148459382, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=CUTLASS-quant=W8A16-routing=Renormalize]": 30.559139130434783, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=DEEPGEMM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.20927576601671, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=FP8_BLOCK_SCALES-routing=Renormalize]": 35.70960465116279, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=NVFP4-routing=Renormalize]": 35.24991267605634, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A16_MXFP4-routing=Renormalize]": 28.97459824046921, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e60_k4_h2048_i1408-seq=1-dtype=torch.bfloat16-backend=TRTLLM-quant=W4A8_NVFP4_FP8-routing=Renormalize]": 58.56761344537815, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.bfloat16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.696147058823527, + "unittest/_torch/modules/moe/test_moe_module.py::test_configurable_moe_single_gpu[e8_k1_h512_i512-seq=8-dtype=torch.float16-backend=CUTLASS-quant=NVFP4-routing=Renormalize]": 21.576210407239817, + "unittest/_torch/modules/test_awq_quantization.py": 20.355017509727627, + "unittest/_torch/modules/test_engram.py": 95.12348051948052, + "unittest/_torch/modules/test_fp4_num_tokens_slice.py": 21.562614084507043, + "unittest/_torch/modules/test_fused_activation_quant.py": 21.362952153110047, + "unittest/_torch/modules/test_fused_add_rms_norm_quant.py": 22.13294797687861, + "unittest/_torch/modules/test_fused_rmsnorm_fp4_quantize.py": 28.106347578347577, + "unittest/_torch/modules/test_gemma4_fused_qkv_prep.py": 25.284024861878454, + "unittest/_torch/modules/test_group_rmn_norm.py": 26.97853846153846, + "unittest/_torch/modules/test_mhc.py": 30.238890876565296, + "unittest/_torch/modules/test_mla_helix.py": 41.2383063583815, + "unittest/_torch/modules/test_moe_host_sharer.py": 22.98750459921156, + "unittest/_torch/modules/test_moe_load_balancer.py": 22.217951550387596, + "unittest/_torch/modules/test_moe_routing.py": 46.00725428571428, + "unittest/_torch/modules/test_rotary_embedding.py": 21.841127167630056, + "unittest/_torch/modules/test_triton_linear.py": 21.558213051823415, + "unittest/_torch/modules/tests_lora_modules": 366.0990338983051, + "unittest/_torch/multi_gpu -m \"not post_merge\"": 274.7847525773196, + "unittest/_torch/multi_gpu_modeling -k \"deepseek\"": 332.8804090909091, + "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp1-bf16-trtllm-deepseekv3_lite]": 83.28676470588235, + "unittest/_torch/multi_gpu_modeling/test_deepseek.py::test_deepseek_streaming[tp4-bf16-trtllm-deepseekv3_lite]": 114.40690045248869, + "unittest/_torch/multimodal -k \"not nemotron_nano_v2_vl_fp8\"": 764.3596734693878, + "unittest/_torch/multimodal/test_mm_encoder_standalone.py -k \"qwen3_30b_a3b_fp8\"": 327.08215211267606, + "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu2\"": 223.64610000000002, + "unittest/_torch/ray_orchestrator/multi_gpu -m \"gpu4\"": 458.21805050505054, + "unittest/_torch/ray_orchestrator/multi_gpu/test_inflight_weight_update.py": 84.21391, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part2\"": 624.9352843137256, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part3\"": 579.3408627450981, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py -m \"part5\"": 577.8696666666666, + "unittest/_torch/ray_orchestrator/multi_gpu/test_llm_update_weights_multi_gpu.py::test_llm_partial_update_weights_qwen35_35b_bf16_tp8_cuda_graph": 210.214125, + "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp1_4instances]": 268.0139898989899, + "unittest/_torch/ray_orchestrator/multi_gpu/test_multi_instance.py::test_multi_instance[tp2_2instances]": 282.0997121212121, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part0\"": 274.29658, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part1\"": 273.73165, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part2\"": 147.88863, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part3\"": 148.11237, + "unittest/_torch/ray_orchestrator/multi_gpu/test_ops.py -m \"part4\"": 204.57628, + "unittest/_torch/ray_orchestrator/single_gpu/test_cache_transceiver_comm.py": 48.72692219020173, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_sleep.py": 112.46935964912281, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part0\"": 711.4720964912281, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part1\"": 661.9107280701754, + "unittest/_torch/ray_orchestrator/single_gpu/test_llm_update_weights.py -m \"part2\"": 486.40483040935675, + "unittest/_torch/sampler -k \"not test_speculative_d2h_parity_real_predictor\"": 631.3018140625001, + "unittest/_torch/sampler/test_beam_search.py": 444.93877317073174, + "unittest/_torch/sampler/test_beam_search_speculative_d2h.py": 127.7736968215159, + "unittest/_torch/sampler/test_logits_logprobs.py": 260.9117658536585, + "unittest/_torch/sampler/test_penalties.py": 56.53724752475248, + "unittest/_torch/sampler/test_token_ban.py": 26.095191964285714, + "unittest/_torch/sampler/test_torch_sampler.py": 226.65034025974023, + "unittest/_torch/sampler/test_trtllm_sampler.py": 92.73561953727507, + "unittest/_torch/speculative/hw_agnostic": 638.9258896551723, + "unittest/_torch/speculative/test_capture_override_leak.py": 20.019795454545456, + "unittest/_torch/speculative/test_eagle3.py": 698.1639762611276, + "unittest/_torch/speculative/test_rejection_buffers_guard.py": 20.62444705882353, + "unittest/_torch/speculative/test_sa_hybrid_state_promotion.py": 20.992551724137932, + "unittest/_torch/test_model_config.py": 25.45066919191919, + "unittest/_torch/test_tensor_lru_cache.py": 25.701808510638298, + "unittest/_torch/test_torch_multi_arange.py": 44.52563636363636, + "unittest/_torch/thop/parallel": 2201.6000647103083, + "unittest/_torch/thop/parallel/test_finegrained_mixed_dtype_gemm.py": 27.07066501240695, + "unittest/_torch/thop/parallel/test_w4a16_linear.py": 21.95623076923077, + "unittest/_torch/thop/parallel/test_w4a8_linear.py": 20.894305210918116, + "unittest/_torch/thop/parallel/test_weight_only_quant_gemm.py": 60.234856079404466, + "unittest/_torch/thop/parallel/test_weight_only_quant_linear.py": 27.418575682382137, + "unittest/_torch/thop/parallel_hw_agnostic": 467.797237012987, + "unittest/_torch/thop/serial": 53.24263645224171, + "unittest/_torch/visual_gen/multi_gpu/test_attn2d_attention.py": 328.0071551724138, + "unittest/_torch/visual_gen/multi_gpu/test_cosmos3_transformer_parallel.py": 210.79964705882352, + "unittest/_torch/visual_gen/multi_gpu/test_flux2_transformer_parallel.py": 233.11, + "unittest/_torch/visual_gen/multi_gpu/test_flux_tp.py": 202.95143220338983, + "unittest/_torch/visual_gen/multi_gpu/test_flux_ulysses.py": 144.93126495726497, + "unittest/_torch/visual_gen/multi_gpu/test_ltx2_async_ulysses.py": 70.44280672268907, + "unittest/_torch/visual_gen/multi_gpu/test_ltx2_ulysses.py": 189.4298151260504, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_attention.py": 44.64089830508475, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_conv.py": 85.23357264957265, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_group_norm.py": 64.13015254237288, + "unittest/_torch/visual_gen/multi_gpu/test_parallel_vae.py": 254.18874789915967, + "unittest/_torch/visual_gen/multi_gpu/test_ring_attention.py": 194.66533898305084, + "unittest/_torch/visual_gen/multi_gpu/test_tp_attention.py": 300.4546120689655, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_async.py": 65.64371186440678, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_attention.py": 371.2901440677966, + "unittest/_torch/visual_gen/multi_gpu/test_ulysses_sage_attention.py": 128.87157627118643, + "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_mapping.py": 214.06044347826085, + "unittest/_torch/visual_gen/multi_gpu/test_visual_gen_multinode.py": 22.233025423728815, + "unittest/_torch/visual_gen/multi_gpu/test_wan_async_ulysses.py": 76.19292372881357, + "unittest/_torch/visual_gen/multi_gpu/test_wan_pipeline_parallel.py": 261.1740084745763, + "unittest/_torch/visual_gen/multi_gpu/test_wan_tp.py": 206.67422807017545, + "unittest/_torch/visual_gen/multi_gpu/test_wan_transformer_parallel.py": 521.5535932203389, + "unittest/_torch/visual_gen/test_attention_cute_dsl.py": 161.75197866666667, + "unittest/_torch/visual_gen/test_attention_cute_dsl_vsa.py": 43.49738057742782, + "unittest/_torch/visual_gen/test_attention_integration.py": 107.21697574123989, + "unittest/_torch/visual_gen/test_attention_perf.py": 219.80059466666668, + "unittest/_torch/visual_gen/test_attention_trtllm_sage.py": 204.94523116883119, + "unittest/_torch/visual_gen/test_cache_dit.py": 170.39281714285715, + "unittest/_torch/visual_gen/test_cosmos3_distilled.py": 36.37447619047619, + "unittest/_torch/visual_gen/test_cosmos3_pipeline.py": 152.43147802197802, + "unittest/_torch/visual_gen/test_cosmos3_transformer.py": 94.60373802816902, + "unittest/_torch/visual_gen/test_flux2_image_conditioning.py": 19.8582, + "unittest/_torch/visual_gen/test_flux_attention.py": 27.198191489361704, + "unittest/_torch/visual_gen/test_flux_infer.py": 25.557968523002423, + "unittest/_torch/visual_gen/test_flux_pipeline.py": 489.605927027027, + "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxCombinedOptimizations::test_all_optimizations_combined": 76.62864705882352, + "unittest/_torch/visual_gen/test_flux_pipeline.py::TestFluxParallelism::test_ulysses_2gpu_correctness": 64.04692796610169, + "unittest/_torch/visual_gen/test_flux_transformer.py": 45.653248677248676, + "unittest/_torch/visual_gen/test_ltx2_attention.py": 24.870206434316355, + "unittest/_torch/visual_gen/test_ltx2_pipeline.py": 727.259218579235, + "unittest/_torch/visual_gen/test_ltx2_transformer.py": 27.478767441860466, + "unittest/_torch/visual_gen/test_model_loader.py": 44.97012464589235, + "unittest/_torch/visual_gen/test_profiler.py": 26.178076923076922, + "unittest/_torch/visual_gen/test_quant_ops.py": 22.4151689373297, + "unittest/_torch/visual_gen/test_qwen_image_infer.py": 25.41444711538462, + "unittest/_torch/visual_gen/test_qwen_image_pipeline.py": 25.53377886977887, + "unittest/_torch/visual_gen/test_teacache.py": 20.578375, + "unittest/_torch/visual_gen/test_tensor_payload.py": 25.7702, + "unittest/_torch/visual_gen/test_trtllm_serve_e2e.py": 493.3113262032085, + "unittest/_torch/visual_gen/test_trtllm_serve_endpoints.py": 29.387322493224932, + "unittest/_torch/visual_gen/test_visual_gen_args.py": 22.076759894459105, + "unittest/_torch/visual_gen/test_visual_gen_params.py": 26.502983565107456, + "unittest/_torch/visual_gen/test_visual_gen_utils.py": 21.9845, + "unittest/_torch/visual_gen/test_wan21_i2v_pipeline.py": 201.90125721784779, + "unittest/_torch/visual_gen/test_wan21_i2v_teacache.py": 332.5180740740741, + "unittest/_torch/visual_gen/test_wan21_t2v_pipeline.py": 183.95970684931507, + "unittest/_torch/visual_gen/test_wan21_t2v_teacache.py": 232.78348850574713, + "unittest/_torch/visual_gen/test_wan21_t2v_teacache_user_coefficients.py": 52.60576533333334, + "unittest/_torch/visual_gen/test_wan22_i2v_pipeline.py": 270.040718579235, + "unittest/_torch/visual_gen/test_wan22_i2v_teacache.py": 138.1834361702128, + "unittest/_torch/visual_gen/test_wan22_t2v_pipeline.py": 256.53519293478263, + "unittest/_torch/visual_gen/test_wan22_t2v_teacache.py": 130.60224316939892, + "unittest/_torch/visual_gen/test_wan22_ti2v_5b_pipeline.py": 123.94177427821522, + "unittest/_torch/visual_gen/test_wan_transformer.py": 77.91112391930837, + "unittest/_torch/visual_gen/test_wan_vae.py": 67.90559340659341, + "unittest/_torch/visual_gen/test_wan_vsa_pipeline.py": 119.16156484149856, + "unittest/_torch/visual_gen/test_warmup.py": 21.650458666666665, + "unittest/_torch/weight_sharing": 25.79753886010363, + "unittest/api_stability": 30.174709183673468, + "unittest/auto_deploy/multigpu/compile": 43.57656435643565, + "unittest/auto_deploy/multigpu/smoke": 154.60278606965176, + "unittest/auto_deploy/multigpu/transformations": 2054.764138613861, + "unittest/auto_deploy/multigpu/transformations/library/test_allreduce_residual_rmsnorm_fusion.py -k \"strategy_auto\"": 121.42506, + "unittest/auto_deploy/multigpu/transformations/library/test_tp_sharding.py::test_moe_tp_shard_nvfp4": 71.84925, + "unittest/auto_deploy/singlegpu/compile": 70.71696694214876, + "unittest/auto_deploy/singlegpu/custom_ops": 525.2752493074793, + "unittest/auto_deploy/singlegpu/custom_ops/attention/test_triton_attention.py::TestSDPADispatch": 39.1398, + "unittest/auto_deploy/singlegpu/custom_ops/mamba/test_flashinfer_mamba_cached_op.py": 48.05372479564033, + "unittest/auto_deploy/singlegpu/custom_ops/moe/test_ad_moe_op.py": 29.51027642276423, + "unittest/auto_deploy/singlegpu/custom_ops/moe/test_trtllm_moe.py": 41.36786178861789, + "unittest/auto_deploy/singlegpu/custom_ops/quantization/test_quant.py": 50.740886178861786, + "unittest/auto_deploy/singlegpu/mlir": 29.758081267217634, + "unittest/auto_deploy/singlegpu/models": 233.6499502074689, + "unittest/auto_deploy/singlegpu/shim": 96.55770539419086, + "unittest/auto_deploy/singlegpu/smoke": 1689.3055637119114, + "unittest/auto_deploy/singlegpu/smoke/test_ad_build_small_single.py -k \"Nemotron-3-Nano-30B-A3B-FP8 or Nemotron-Nano-3-30B-A3.5B-dev or Llama-4-Scout\"": 249.0911842818428, + "unittest/auto_deploy/singlegpu/smoke/test_ad_speculative_decoding.py": 209.38515135135134, + "unittest/auto_deploy/singlegpu/transformations": 333.13563862068963, + "unittest/auto_deploy/singlegpu/transformations/library/test_fuse_relu2_quant_nvfp4.py": 37.48995640326975, + "unittest/auto_deploy/singlegpu/transformations/library/test_moe_fusion.py": 66.10721253405994, + "unittest/auto_deploy/singlegpu/transformations/library/test_nvfp4_swiglu.py": 34.6551825613079, + "unittest/auto_deploy/singlegpu/transformations/library/test_quant_fusion.py": 40.50969945355191, + "unittest/auto_deploy/singlegpu/transformations/library/test_quant_moe.py": 32.301112021857925, + "unittest/auto_deploy/singlegpu/transformations/library/test_quantization.py": 33.53743989071038, + "unittest/auto_deploy/singlegpu/transformations/library/test_trtllm_gen_diag.py": 41.74485792349727, + "unittest/auto_deploy/singlegpu/utils": 45.690824827586205, + "unittest/auto_deploy/standalone": 749.0343914373088, + "unittest/bindings": 33.233741379310345, + "unittest/disaggregated/region/test_aux.py": 20.118003086419755, + "unittest/disaggregated/region/test_block.py": 23.155147471910112, + "unittest/disaggregated/region/test_page.py": 20.19467365269461, + "unittest/disaggregated/region/test_region.py": 20.12785579937304, + "unittest/disaggregated/test_agent_multi_backends.py": 82.44263218390805, + "unittest/disaggregated/test_bounce.py": 23.681536211699164, + "unittest/disaggregated/test_cache_reuse_adapter.py": 25.694189973614776, + "unittest/disaggregated/test_cache_transceiver_harness.py": 54.7884509202454, + "unittest/disaggregated/test_cache_transceiver_harness_report.py": 22.569489296636085, + "unittest/disaggregated/test_cache_transceiver_precheck_e2e.py": 102.7121677852349, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and no_window\"": 263.90550151975685, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v1 and uniform_window\"": 232.48220388349515, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and no_window\"": 320.16261075949365, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and uniform_window\"": 325.7727388535032, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver -k \"v2 and vswa\"": 298.592413897281, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v1\"": 63.74947222222222, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_boundary_lengths -k \"v2\"": 79.69995151515151, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"no_window\"": 203.92816666666667, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"uniform_window\"": 208.466928125, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_uneven_pp -k \"v2 and vswa\"": 127.31946202531645, + "unittest/disaggregated/test_cache_transceiver_single_process.py::test_cache_transceiver_v1_dsa_indexer": 54.168414110429445, + "unittest/disaggregated/test_cluster_storage.py": 47.79166496163683, + "unittest/disaggregated/test_coordinator_e2e.py": 72.31078260869565, + "unittest/disaggregated/test_coordinator_worker.py": 32.91763541666666, + "unittest/disaggregated/test_deepseek_v4_kv_transfer.py": 456.1797496206373, + "unittest/disaggregated/test_disagg_cluster_manager_worker.py": 56.201279898218836, + "unittest/disaggregated/test_disagg_openai_client.py": 33.44577353689567, + "unittest/disaggregated/test_disagg_utils.py": 23.457208333333334, + "unittest/disaggregated/test_disaggregated_params.py": 20.980890675241156, + "unittest/disaggregated/test_extractor.py": 27.114717696629214, + "unittest/disaggregated/test_kv_transfer.py": 289.1068039215686, + "unittest/disaggregated/test_kv_transfer_mp.py": 84.64788253012048, + "unittest/disaggregated/test_mamba_transfer.py": 34.03104336734694, + "unittest/disaggregated/test_messenger.py": 23.249504201680672, + "unittest/disaggregated/test_minimax_m3_kv_transfer.py": 362.53847222222225, + "unittest/disaggregated/test_openai_disagg_server.py": 27.016676020408163, + "unittest/disaggregated/test_openai_disagg_service.py": 33.99901526717557, + "unittest/disaggregated/test_openai_server_info.py": 25.25883333333333, + "unittest/disaggregated/test_peer.py": 23.113104285714286, + "unittest/disaggregated/test_perf_logger.py": 20.46639936102236, + "unittest/disaggregated/test_pool_matching.py": 20.040252427184466, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first and not ctx_first_sync\"": 291.7294591836735, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"ctx_first_sync\"": 289.61003000000005, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first1\"": 293.302887755102, + "unittest/disaggregated/test_py_cache_transceiver_mp.py -k \"gen_first2\"": 316.01223, + "unittest/disaggregated/test_rank_info.py": 20.087990936555894, + "unittest/disaggregated/test_remoteDictionary.py": 28.16403486750349, + "unittest/disaggregated/test_request_id.py": 20.111296072507553, + "unittest/disaggregated/test_router.py": 27.52228693181818, + "unittest/disaggregated/test_sync_transfer_timing.py": 20.093430420711975, + "unittest/disaggregated/test_transceiver_bounded_polling.py": 20.062371335504885, + "unittest/dynamo": 31.191930591259638, + "unittest/executor/test_base_worker.py": 132.89819083969468, + "unittest/executor/test_event_loop_error_broadcast.py": 23.20425503355705, + "unittest/executor/test_fatal_error_health_check.py": 27.783138539042824, + "unittest/executor/test_ipc.py": 28.95289724310777, + "unittest/executor/test_multi_frontend_routing.py": 20.487352668213457, + "unittest/executor/test_postprocessor_hook.py": 29.449583979328168, + "unittest/executor/test_proxy_fast_death.py": 30.67723711340206, + "unittest/executor/test_proxy_postproc_terminate.py": 25.31334010152284, + "unittest/executor/test_rpc.py": 434.0945229357798, + "unittest/executor/test_stats_serializer.py": 21.669697547683924, + "unittest/inputs/test_chat_template_dispatch.py": 26.911028645833333, + "unittest/inputs/test_content_format.py": 25.583408268733848, + "unittest/inputs/test_multimodal.py": 25.46571794871795, + "unittest/inputs/test_multimodal_input_processor.py": 25.350826873385014, + "unittest/inputs/test_url_validation.py": 25.648604651162792, + "unittest/inputs/test_video_decode.py": 25.576173126614986, + "unittest/kv_cache_manager_v2_tests": 228.90979301075268, + "unittest/llmapi/apps/_test_openai_chat_multimodal.py::test_single_chat_session_image_embeds -m needs_l40s": 157.64357537688443, + "unittest/llmapi/apps/_test_openai_embeddings.py": 154.04824111675129, + "unittest/llmapi/apps/_test_openai_multi_gpu.py -m \"part0\"": 121.33642857142857, + "unittest/llmapi/apps/test_chat_utils.py": 27.294929824561404, + "unittest/llmapi/apps/test_disagg_serving_perf_metrics.py": 80.77289215686274, + "unittest/llmapi/apps/test_encode_batcher.py": 27.280994923857868, + "unittest/llmapi/apps/test_harmony_channel_validation.py": 30.592439086294416, + "unittest/llmapi/apps/test_tool_parsers.py": 29.554723618090453, + "unittest/llmapi/test_additional_model_outputs.py -m \"gpu1\"": 62.16573945409429, + "unittest/llmapi/test_async_llm.py -m \"gpu2\"": 82.15391000000001, + "unittest/llmapi/test_async_llm.py -m \"gpu4\"": 88.93448484848484, + "unittest/llmapi/test_async_llm.py -m \"not (gpu2 or gpu4)\"": 196.92821637426903, + "unittest/llmapi/test_bench_async.py": 30.84113043478261, + "unittest/llmapi/test_deepseek_v4_tokenizer.py": 20.848953623188404, + "unittest/llmapi/test_embedding_arch_routing.py": 26.84065891472868, + "unittest/llmapi/test_executor.py": 29.101611111111108, + "unittest/llmapi/test_gc_utils.py": 31.859434343434344, + "unittest/llmapi/test_kv_cache_dtype_override.py": 26.816649122807018, + "unittest/llmapi/test_llm.py -m \"not part0\"": 225.55044444444445, + "unittest/llmapi/test_llm.py -m \"part0\"": 327.5163333333333, + "unittest/llmapi/test_llm_args.py": 132.44800253807108, + "unittest/llmapi/test_llm_encode.py": 64.248390625, + "unittest/llmapi/test_llm_encode_multi_item.py": 129.08888511749348, + "unittest/llmapi/test_llm_kv_cache_events.py::test_llm_api_attention_dp_kv_events": 54.56314285714286, + "unittest/llmapi/test_llm_multi_gpu_pytorch.py -m \"gpu4\"": 78.35294554455446, + "unittest/llmapi/test_llm_pytorch.py -m \"part0\"": 923.5839957507083, + "unittest/llmapi/test_llm_pytorch.py -m \"part1\"": 257.2634564315353, + "unittest/llmapi/test_llm_pytorch.py -m \"part2\"": 480.78776657458565, + "unittest/llmapi/test_llm_pytorch.py -m \"part3\"": 275.2785779310345, + "unittest/llmapi/test_llm_pytorch.py::test_nemotron_nas_lora": 247.856625, + "unittest/llmapi/test_llm_quant.py": 21.912051169590644, + "unittest/llmapi/test_llm_telemetry.py": 170.3385606060606, + "unittest/llmapi/test_llm_telemetry.py::TestTelemetryArchitectureExtraction": 71.03602319587628, + "unittest/llmapi/test_llm_telemetry.py::TestTelemetryPyTorchBackend": 71.8847385786802, + "unittest/llmapi/test_llm_utils.py": 30.531091836734692, + "unittest/llmapi/test_memory_profiling.py::test_pyexecutor_and_kvcache_share_execution_stream": 44.25662077922078, + "unittest/llmapi/test_mpi_session.py": 160.02251399491092, + "unittest/llmapi/test_mpi_session.py::test_llmapi_launch_multiple_tasks": 87.06310416666668, + "unittest/llmapi/test_reasoning_parser.py": 30.21788888888889, + "unittest/llmapi/test_request_priority.py": 25.484827067669173, + "unittest/llmapi/test_rlhf_utils.py": 25.711772727272727, + "unittest/llmapi/test_serialization.py": 25.378102301790282, + "unittest/llmapi/test_utils.py": 28.242190839694658, + "unittest/metrics/test_collector.py": 24.564216438356166, + "unittest/models/test_quant_config_utils.py": 25.456101265822785, + "unittest/others/test_cache_transceiver_precheck_config.py": 30.594468208092486, + "unittest/others/test_cache_transceiver_precheck_run.py": 27.6636449704142, + "unittest/others/test_convert_utils.py": 25.22678756476684, + "unittest/others/test_http_utils_fail_fast.py": 29.93428426395939, + "unittest/others/test_kv_cache_transceiver.py::test_async_transfer_keeps_llm_request_alive": 12.767763144717076, + "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mha]": 27.682636863136864, + "unittest/others/test_kv_cache_transceiver.py::test_cancel_request_in_transmission[mla]": 16.382363386613385, + "unittest/others/test_kv_cache_transceiver.py::test_cpp_nixl_sync_transfer_stress": 29.801800862068966, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[NIXL-mha-ctx_fp16_gen_fp16]": 23.806841658341657, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mha-ctx_fp16_gen_fp16]": 16.071417549167926, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[PYTHON-mla-ctx_fp16_gen_fp16]": 15.999426494345718, + "unittest/others/test_kv_cache_transceiver.py::test_kv_cache_transceiver_single_process[UCX-mha-ctx_fp16_gen_fp16]": 12.542696803196803, + "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_silent_when_unset": 10.897987981972959, + "unittest/others/test_kv_cache_transceiver.py::test_kv_transfer_timeout_warns_once_per_request": 10.896794441662495, + "unittest/others/test_lm_eval.py": 28.92114, + "unittest/others/test_lora_manager.py": 35.860937984496125, + "unittest/others/test_lora_module_count.py": 27.002921319796954, + "unittest/others/test_time_breakdown.py": 27.481543589743588, + "unittest/others/test_tracing.py": 144.11626424870465, + "unittest/quantization/test_mode.py": 25.516699745547072, + "unittest/scaffolding": 395.9213879781421, + "unittest/scripts": 31.20939588688946, + "unittest/tools": 386.9344300518135, + "unittest/tools/test_layer_wise_benchmarks.py::test_deepseek_r1_ctx_dep[1]": 20.868882183908045, + "unittest/tools/test_layer_wise_benchmarks.py::test_nemotron_gen_dep[1]": 103.46749022164276, + "unittest/tools/test_layer_wise_benchmarks.py::test_performance_alignment[1]": 27.55558333333333, + "unittest/tools/test_layer_wise_benchmarks.py::test_qwen3_next_gen_tep[1]": 84.47367901234567, + "unittest/usage/test_collectors.py": 23.337350631136044, + "unittest/usage/test_config.py": 22.926307905686546, + "unittest/usage/test_e2e_capture.py": 22.975632911392402, + "unittest/usage/test_llmapi_config_telemetry_docs.py": 27.81584223918575, + "unittest/usage/test_opt_out.py": 22.998687065368568, + "unittest/usage/test_reporter.py": 26.33069496402878, + "unittest/usage/test_schema.py": 23.32576157082749, + "unittest/usage/test_transport.py": 29.48707735583685, + "unittest/utils/test_logger.py": 25.696051813471502, + "unittest/utils/test_util.py": 28.391819796954316, + "unittest/visual_gen/test_iteration_stats.py": 25.375496368038743, + "unittest/visual_gen/test_media_encoding.py": 25.67736519607843, + "unittest/visual_gen/test_output.py": 25.828207637231504, + "verl/test_verl_cases.py::test_async_generate": 681.7801428571429, + "verl/test_verl_cases.py::test_async_memory_management": 128.481, + "verl/test_verl_cases.py::test_init_without_device_mesh": 163.0692857142857, + "verl/test_verl_cases.py::test_inter_node_trtllm_rollout": 184.09014285714287, + "verl/test_verl_cases.py::test_make_async_request_get_method": 47.95542857142857, + "verl/test_verl_cases.py::test_make_async_request_http_error": 51.29528571428572, + "verl/test_verl_cases.py::test_make_async_request_max_attempts_exceeded": 50.296, + "verl/test_verl_cases.py::test_make_async_request_post_method": 45.08557142857143, + "verl/test_verl_cases.py::test_multimodal_different_image_sizes": 265.54571428571427, + "verl/test_verl_cases.py::test_multimodal_generate_with_image": 254.34057142857142, + "verl/test_verl_cases.py::test_multimodal_text_only_fallback": 215.1765714285714, + "verl/test_verl_cases.py::test_placement_group_multi_node_multi_replica": 48.83014285714285, + "verl/test_verl_cases.py::test_placement_group_multi_node_ray_resource_pool": 42.849, + "verl/test_verl_cases.py::test_placement_group_with_ray_resource_pool": 42.37528571428572, + "verl/test_verl_cases.py::test_placement_group_with_sub_ray_resource_pool": 41.65785714285715, + "verl/test_verl_cases.py::test_unimodal_batch_generate": 236.276, + "verl/test_verl_cases.py::test_unimodal_generate": 229.85028571428572, + "verl/test_verl_cases.py::test_wake_sleep_cycle": 193.665, + "visual_gen/test_visual_gen_benchmark.py::test_offline_benchmark": 90.47928571428571, + "visual_gen/test_visual_gen_benchmark.py::test_online_benchmark[openai-videos]": 91.72266666666667 } diff --git a/tests/integration/test_lists/waives.txt b/tests/integration/test_lists/waives.txt index b24098d34425..473eb870caf4 100644 --- a/tests/integration/test_lists/waives.txt +++ b/tests/integration/test_lists/waives.txt @@ -171,6 +171,7 @@ full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/Qwen3-235B-A22B-tp16-mmlu] SK full:B200/test_e2e.py::test_multi_nodes_eval[Qwen3/saved_models_Qwen3-235B-A22B_nvfp4_hf-tp16-mmlu] SKIP (https://nvbugs/6424188) full:B200/test_e2e.py::test_qwen_e2e_cpprunner_large_new_tokens[DeepSeek-R1-Distill-Qwen-1.5B-DeepSeek-R1-Distill-Qwen-1.5B] SKIP (https://nvbugs/6414760) full:B300/accuracy/test_disaggregated_serving.py::TestDeepSeekV3Lite::test_auto_dtype_with_helix[fifo_v2-cudagraph:with_padding-pp1tp2cp2] SKIP (https://nvbugs/6410881) +full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[bf16-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6539942) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronNanoV3::test_accuracy[fp8-4-attn_dp_off-trtllm] SKIP (https://nvbugs/6483370) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-flashinfer] SKIP (https://nvbugs/6478645) full:B300/accuracy/test_llm_api_autodeploy.py::TestNemotronSuperV3::test_mtp[bf16_ws4_180gb-trtllm] SKIP (https://nvbugs/6478645) From e6e7f0b6e08ed248ed726188ace6e49fb3c32b46 Mon Sep 17 00:00:00 2001 From: Asfiya Baig Date: Wed, 5 Aug 2026 15:40:00 -0700 Subject: [PATCH 14/14] fix rebase bugs Signed-off-by: Asfiya Baig --- tests/integration/defs/.test_durations | 1 - 1 file changed, 1 deletion(-) diff --git a/tests/integration/defs/.test_durations b/tests/integration/defs/.test_durations index 6d7ec789d334..bb0f8163487f 100644 --- a/tests/integration/defs/.test_durations +++ b/tests/integration/defs/.test_durations @@ -646,7 +646,6 @@ "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_bf16_empty_batch[DeepSeek-V3-Lite-bf16]": 190.4600707070707, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_attention_dp[DeepSeek-V3-Lite-fp8]": 128.8671616161616, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_nixl[DeepSeek-V3-Lite-fp8]": 121.09470707070707, - "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_transceiver_runtime_python[DeepSeek-V3-Lite-fp8]": 118.99839393939394, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]": 99.04030303030302, "disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx_tp1_single_gpu[DeepSeek-V3-Lite-fp8]": 136.5389226006192, "disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0]": 72.53091237113402,