Skip to content

[https://nvbugs/6185173][fix] Set mamba ssm cache to fp32 for NemotronV2 #14448

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

Merged
nvchenghaoz merged 2 commits into from
May 28, 2026
Merged

[https://nvbugs/6185173][fix] Set mamba ssm cache to fp32 for NemotronV2 #14448

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
09d8ca3
[nvbugs/6185173][fix] AutoDeploy: use float32 SSM cache for Nemotron-…
tensorrt-cicd May 22, 2026
d3e52b9
Update nemotron-nano-9b-v2.yaml
nvchenghaoz May 27, 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
3 changes: 3 additions & 0 deletions examples/auto_deploy/model_registry/configs/nemotron-nano-9b-v2.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,9 @@ trust_remote_code: true
kv_cache_config:
enable_block_reuse: false
free_gpu_memory_fraction: 0.7
# H20 shows an MMLU accuracy drop with the default BF16 recurrent SSM cache.
# Keep SSM state in FP32 for stable Triton SSM generation.
mamba_ssm_cache_dtype: float32

# Keep max_batch_size as in the PyTorch test to avoid OOM
max_batch_size: 128
Expand Down
3 changes: 0 additions & 3 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -279,9 +279,6 @@ full:GH200/examples/test_nemotron.py::test_llm_nemotron_3_8b_1gpu[bfloat16-fp8]
full:GH200/examples/test_qwen2audio.py::test_llm_qwen2audio_single_gpu[qwen2_audio_7b_instruct] SKIP (arm is not supported)
full:GH200/unittest/trt/model_api/test_model_quantization.py SKIP (https://nvbugs/4979955)
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_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[False] SKIP (https://nvbugs/6185173)
full:H20/accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_auto_dtype[True] SKIP (https://nvbugs/6185173)
full:H20/accuracy/test_llm_api_autodeploy.py::TestNemotronV2::test_fp8[True] SKIP (https://nvbugs/6185173)
full:H20/accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_chunked_prefill[triton-auto] SKIP (https://nvbugs/6026676)
full:RTX/accuracy/test_llm_api_pytorch.py::TestGemma3_1BInstruct::test_auto_dtype SKIP (https://nvbugs/5569696)
full:RTXPro6000D/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] SKIP (https://nvbugs/5948435)
Expand Down
Loading