Skip to content

[https://nvbugs/5522462][fix] Fix FP8 scout illegal memory access #7845

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
merged 1 commit into from
Sep 19, 2025
Merged

[https://nvbugs/5522462][fix] Fix FP8 scout illegal memory access #7845

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 2 additions & 1 deletion ...rrt_llm/kernels/decoderMaskedMultiheadAttention/decoderMaskedMultiheadAttentionTemplate.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1363,7 +1363,8 @@ __global__ void __launch_bounds__(MAX_THEADS_PER_BLOCK, MIN_BLOCKS_PER_SM) maske
#ifndef MMHA_USE_FP32_ACCUM_FOR_LOGITS
if (sizeof(Tk) != 4)
{
auto const max_timesteps = min(timestep, static_cast<unsigned>(cyclic_kv_cache_len));
auto const max_timesteps
= min(timestep, min(chunked_attention_size, static_cast<unsigned>(cyclic_kv_cache_len)));
logits_smem_ += divUp(max_timesteps + 1, 4u) * 16;
}
Tk* logits_smem = reinterpret_cast<Tk*>(logits_smem_);
Expand Down
11 changes: 11 additions & 0 deletions tests/integration/test_lists/test-db/l0_dgx_b200.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,18 @@ l0_dgx_b200:
- accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[tep4_latency_moe_trtllm-torch_compile=True]
- accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_trtllm-torch_compile=False]
- accuracy/test_llm_api_pytorch.py::TestQwen3_30B_A3B::test_nvfp4[dep4_latency_moe_cutlass-torch_compile=False]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8-cuda_graph=False]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8ep4-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp8ep8-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp4-cuda_graph=False]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp4ep2-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_auto_dtype[tp4ep4-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8[tp8ep8-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8[tp4-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp4[tp8ep8-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp4[tp4-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8_chunked_prefill[tp4ep4-cuda_graph=True]
- accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp4_chunked_prefill[tp4ep4-cuda_graph=True]
- disaggregated/test_disaggregated.py::test_disaggregated_deepseek_v3_lite_fp8_ucx[DeepSeek-V3-Lite-fp8]
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[tp4-trtllm-auto]
- accuracy/test_llm_api_pytorch.py::TestGPTOSS::test_w4_4gpus[ep4-cutlass-auto]
Expand Down
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 @@ -337,7 +337,6 @@ full:H100/accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_fp8
full:H100/accuracy/test_llm_api_pytorch.py::TestLlama4MaverickInstruct::test_fp8_eagle3[tp8-torch_compile=True] SKIP (https://nvbugs/5483534)
full:A100/test_e2e.py::test_ptp_quickstart_multimodal[NVILA-8B-FP16-vila/NVILA-8B-video-False] SKIP (https://nvbugs/5453725)
test_e2e.py::test_ptp_scaffolding[DeepSeek-R1-Distill-Qwen-7B-DeepSeek-R1/DeepSeek-R1-Distill-Qwen-7B] SKIP (https://nvbugs/5517260)
accuracy/test_llm_api_pytorch.py::TestLlama4ScoutInstruct::test_fp8_chunked_prefill[tp4ep4-cuda_graph=True] SKIP (https://nvbugs/5522462)
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] SKIP (https://nvbugs/5522746)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV3Lite::test_nvfp4_4gpus[moe_backend=CUTLASS-mtp_nextn=0-tp2pp2-fp8kv=False-attention_dp=False-cuda_graph=False-overlap_scheduler=False-torch_compile=False] SKIP (https://nvbugs/5522746)
test_e2e.py::test_ptp_quickstart_multimodal[NVILA-8B-FP16-vila/NVILA-8B-image-False] SKIP (https://nvbugs/5523925)
Expand Down