Skip to content

[TRTLLM-15076][test] Remove Bielik-11B-v2.2-Instruct an… #17345

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
xinhe-nv wants to merge 4 commits into
base: main
Choose a base branch
from
Open

[TRTLLM-15076][test] Remove Bielik-11B-v2.2-Instruct an… #17345

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5837f80
[TRTLLM-15076][TRTLLM-15077][test] Remove Bielik-11B-v2.2-Instruct an…
xinhe-nv Aug 6, 2026
d5753fc
Merge branch 'main' into prune
xinhe-nv Aug 6, 2026
d9bc9b6
Merge branch 'main' into prune
xinhe-nv Aug 6, 2026
7300624
Merge branch 'main' into prune
xinhe-nv Aug 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 0 additions & 2 deletions tests/integration/defs/accuracy/references/gsm8k.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -293,8 +293,6 @@ nvidia/Llama-3_1-Nemotron-Ultra-253B-v1:
- quant_algo: FP8
kv_cache_quant_algo: FP8
accuracy: 94.16
kanana-1.5-2.1b-instruct-2505:
- accuracy: 75.81
google/gemma-3-1b-it:
- accuracy: 25.52 # score getting from lm-eval with HF implementation
- quant_algo: FP8
Expand Down
2 changes: 0 additions & 2 deletions tests/integration/defs/accuracy/references/mmlu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -334,8 +334,6 @@ nvidia/Llama-3_1-Nemotron-Ultra-253B-v1:
- quant_algo: FP8
kv_cache_quant_algo: FP8
accuracy: 83.36
kanana-1.5-2.1b-instruct-2505:
- accuracy: 56.89
mistralai/Ministral-8B-Instruct-2410:
- accuracy: 66.35
- quant_algo: FP8
Expand Down
17 changes: 0 additions & 17 deletions tests/integration/defs/accuracy/test_llm_api_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5152,23 +5152,6 @@ def test_auto_dtype(self):
task.evaluate(llm)


class TestKanana_Instruct(LlmapiAccuracyTestHarness):

@BowenFu BowenFu Aug 6, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

TestKanana_Instruct is the only coverage for an in-flight product fix. #14707 (open) fixes the very bug this test is waived under — nvbugs/6209806 — by relaxing transformers 5.x LlamaConfig.validate_architecture in _torch/pyexecutor/config_utils.py and tokenizer/tokenizer.py so models with an explicit head_dim load. It removes this exact waives.txt line to re-enable this test, and ships no replacement test of its own.

So if this lands first, that fix merges with zero regression coverage for the load path it changes.

Fine either way once it's a decision rather than a side effect — if kanana is genuinely retired, say so on #14707 so it gets a focused config_utils unit test instead; if it isn't, keep this class and let #14707 do the unwaive. The waives.txt overlap itself is not a problem (three-way merge is clean).

Everything else here checks out: no dangling references left for either model, check_test_list.py --validate passes, the bielik_11b_v2.2_instruct alias is removed from _model_paths.py too (a real retirement, not just a coverage drop), and no import in test_llm_pytorch.py is orphaned.

@xinhe-nv xinhe-nv Aug 6, 2026

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@BowenFu should we keep this model or not?

MODEL_NAME = "kanana-1.5-2.1b-instruct-2505"
MODEL_PATH = f"{llm_models_root()}/kanana-1.5-2.1b-instruct-2505"

@pytest.mark.skip_device_not_contain(["H20", "H100"])
def test_auto_dtype(self):
"RCCA: https://nvbugspro.nvidia.com/bug/5310520"
pytorch_config = dict(cuda_graph_config=CudaGraphConfig(
enable_padding=True, max_batch_size=384))
with LLM(self.MODEL_PATH, **pytorch_config,
enable_attention_dp=True) as llm:
task = MMLU(self.MODEL_NAME)
task.evaluate(llm)
task = GSM8K(self.MODEL_NAME)
task.evaluate(llm)


@skip_pre_hopper
@pytest.mark.skip_less_device_memory(80000)
class TestGPTOSS(LlmapiAccuracyTestHarness):
Expand Down
1 change: 0 additions & 1 deletion tests/integration/defs/perf/_model_paths.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,6 @@
"phi_4_multimodal_instruct": "multimodals/Phi-4-multimodal-instruct",
"phi_4_multimodal_instruct_fp4": "multimodals/Phi-4-multimodal-instruct-FP4",
"phi_4_multimodal_instruct_fp8": "multimodals/Phi-4-multimodal-instruct-FP8",
"bielik_11b_v2.2_instruct": "Bielik-11B-v2.2-Instruct",
"gpt_oss_120b_fp4": "gpt_oss/gpt-oss-120b",
"gpt_oss_20b_fp4": "gpt_oss/gpt-oss-20b",
"gpt_oss_120b_eagle3": "gpt_oss/gpt-oss-120b",
Expand Down
1 change: 0 additions & 1 deletion tests/integration/test_lists/qa/llm_function_core.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -538,7 +538,6 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[trtllm-aut
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[trtllm-fp8]
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-auto]
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4a16[dp4-fp8]
accuracy/test_llm_api_pytorch.py::TestKanana_Instruct::test_auto_dtype
accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[dep8]
accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[ep8]
accuracy/test_llm_api_pytorch.py::TestKimiK25::test_nvfp4[tp8]
Expand Down
15 changes: 0 additions & 15 deletions tests/integration/test_lists/test-db/l0_perf.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,18 +14,3 @@ l0_perf:
backend: pytorch
tests:
- perf/test_perf.py::test_perf[llama_v3.1_8b_instruct-bench-pytorch-float16-input_output_len:128,128-reqs:8192]

- condition:
ranges:
system_gpu_count:
gte: 1
lte: 1
wildcards:
gpu:
- '*l40s*'
linux_distribution_name: ubuntu*
terms:
stage: pre_merge
backend: pytorch
tests:
- perf/test_perf.py::test_perf[bielik_11b_v2.2_instruct-bench-pytorch-bfloat16-input_output_len:128,128]
1 change: 0 additions & 1 deletion tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,6 @@ accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-dp4-cutl
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-ep4-cutlass-auto] SKIP (https://nvbugs/5596343)
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[v2_kv_cache-tp4-cutlass-auto] SKIP (https://nvbugs/5596343)
accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[cutlass-auto] SKIP (https://nvbugs/5596343)
accuracy/test_llm_api_pytorch.py::TestKanana_Instruct::test_auto_dtype SKIP (https://nvbugs/6209806)
accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[pp4-attn_backend=TRTLLM-torch_compile=False] SKIP (https://nvbugs/6490043)
accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp2pp2-attn_backend=FLASHINFER-torch_compile=True] SKIP (https://nvbugs/6422337)
accuracy/test_llm_api_pytorch.py::TestLlama3_1_8BInstruct::test_bfloat16_4gpus[tp4-attn_backend=TRTLLM-torch_compile=False] SKIP (https://nvbugs/5616182)
Expand Down
53 changes: 0 additions & 53 deletions tests/unittest/llmapi/test_llm_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -726,59 +726,6 @@ def test_llama_3_3_70b_fp8_with_squad_lora_tp2() -> None:
llm.shutdown()


@skip_gpu_memory_less_than_80gb
@pytest.mark.part2
@test_lora_with_and_without_cuda_graph
def test_bielik_11b_v2_2_instruct_multi_lora(cuda_graph_config) -> None:
model_dir = f"{llm_models_root()}/Bielik-11B-v2.2-Instruct"

target_modules = ['attn_q', 'attn_k', 'attn_v']

# Set up temporary directory for LoRA adapters
with tempfile.TemporaryDirectory() as lora_dir:
print("Creating dummy LoRAs...")

model = AutoModelForCausalLM.from_pretrained(model_dir,
dtype=torch.bfloat16,
device_map="auto")
hf_modules = ["q_proj", "k_proj", "v_proj"]
peft_lora_config = PeftLoraConfig(r=8,
target_modules=hf_modules,
bias="none",
task_type="CAUSAL_LM")
lora_paths = []
for i in range(2):
lora_model = get_peft_model(model, peft_lora_config)
for param in lora_model.parameters():
param.data.zero_()
lora_path = f"{lora_dir}/lora_{i}"
lora_model.save_pretrained(lora_path)
lora_paths.append(lora_path)

trtllm_lora_config = LoraConfig(lora_target_modules=target_modules,
max_lora_rank=8,
max_loras=2,
max_cpu_loras=2)
llm = LLM(model_dir,
lora_config=trtllm_lora_config,
cuda_graph_config=cuda_graph_config)

prompts = [
"Kim był Mikołaj Kopernik i z czego zasłynął?",
"Gdzie znajduje się stolica Polski?",
]
lora_req1 = LoRARequest("lora-1", 0, lora_paths[0])
lora_req2 = LoRARequest("lora-2", 1, lora_paths[1])
lora_requests = [lora_req1, lora_req2]
sampling_params = SamplingParams(max_tokens=200)

outputs = llm.generate(prompts,
sampling_params,
lora_request=lora_requests)

assert len(outputs) == 2


@pytest.mark.part2
@test_lora_with_and_without_cuda_graph
def test_gemma3_1b_instruct_multi_lora(cuda_graph_config) -> None:
Expand Down
Loading