From 000fbef98a1e9a2ca4e394f9be4cc03f28c29bba Mon Sep 17 00:00:00 2001 From: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com> Date: Thu, 29 Oct 2026 12:27:59 -0700 Subject: [PATCH] Bump CI Megatron-Bridge to e42042fa8; test Qwen3.5-VL bridge export with full NVFP4 Megatron-Bridge e42042fa8 (NVIDIA-NeMo/Megatron-Bridge#6714) exports quantized GatedDeltaNet in_proj, so the bridge-vs-Megatron-exporter parity test now covers Qwen3.5-VL-MoE with every linear layer in NVFP4 instead of MLP only. Co-Authored-By: Claude Opus 5.5 Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com> --- .github/actions/megatron-from-source/action.yml | 2 +- examples/megatron_bridge/README.md | 2 +- .../torch/export/test_quantized_mbridge_export.py | 6 +++--- 3 files changed, 5 insertions(+), 5 deletions(-) diff --git a/.github/actions/megatron-from-source/action.yml b/.github/actions/megatron-from-source/action.yml index d8d94592cb1..77e95b4cbc6 100644 --- a/.github/actions/megatron-from-source/action.yml +++ b/.github/actions/megatron-from-source/action.yml @@ -6,7 +6,7 @@ description: >- inputs: commit: description: Megatron-Bridge commit. Keep in sync with examples/megatron_bridge/README.md. - default: 0652ea797ab629e6451c64e1c92f976501b3b8b3 + default: e42042fa83aed837ce6bc13480c06350ea091f1a runs: using: composite diff --git a/examples/megatron_bridge/README.md b/examples/megatron_bridge/README.md index c16a5c6e147..a1130e2f4ce 100644 --- a/examples/megatron_bridge/README.md +++ b/examples/megatron_bridge/README.md @@ -51,7 +51,7 @@ Some features (e.g. [`--vllm_fakequant_export`](#evaluating-a-qad-checkpoint-in- ```bash export MBRIDGE_DIR=${PWD}/Megatron-Bridge git clone https://github.com/NVIDIA-NeMo/Megatron-Bridge.git ${MBRIDGE_DIR} -git -C ${MBRIDGE_DIR} checkout 0652ea797ab629e6451c64e1c92f976501b3b8b3 +git -C ${MBRIDGE_DIR} checkout e42042fa83aed837ce6bc13480c06350ea091f1a git -C ${MBRIDGE_DIR} submodule update --init 3rdparty/Megatron-LM ``` diff --git a/tests/gpu_megatron/torch/export/test_quantized_mbridge_export.py b/tests/gpu_megatron/torch/export/test_quantized_mbridge_export.py index 9cfa1549835..f74b917ff69 100644 --- a/tests/gpu_megatron/torch/export/test_quantized_mbridge_export.py +++ b/tests/gpu_megatron/torch/export/test_quantized_mbridge_export.py @@ -101,11 +101,11 @@ def bridge_workers(request): [ (partial(create_tiny_llama_dir, with_tokenizer=True), "nvfp4"), (partial(create_tiny_qwen3_dir, with_tokenizer=True), "fp8"), - # The bridge can't export quantized Mamba or GatedDeltaNet projections yet. + # The bridge can't export quantized Mamba projections yet. (partial(create_tiny_nemotron_h_dir, with_tokenizer=True), "nvfp4_mlp_only"), - (partial(create_tiny_qwen3_5_moe_vl_dir, with_processor=True), "nvfp4_mlp_only"), + (partial(create_tiny_qwen3_5_moe_vl_dir, with_processor=True), "nvfp4"), ], - ids=["llama-nvfp4", "qwen3-fp8", "nemotron_h-nvfp4_mlp_only", "qwen3_5_moe_vl-nvfp4_mlp_only"], + ids=["llama-nvfp4", "qwen3-fp8", "nemotron_h-nvfp4_mlp_only", "qwen3_5_moe_vl-nvfp4"], ) def test_matches_megatron_exporter( dist_workers_size_1, bridge_workers, tmp_path, create_source, quant_cfg_name