Skip to content

[https://nvbugs/6186880][fix] In deep_ep.py, fall back to the pre-quant dispatch path when hidden_states_sf is #14404

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
hyukn merged 2 commits into from
May 26, 2026
Merged

[https://nvbugs/6186880][fix] In deep_ep.py, fall back to the pre-quant dispatch path when hidden_states_sf is #14404

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
75febe4
[nvbugs/6186880][fix] Handle non-quantized MoE in DeepEP post-quant d…
tensorrt-cicd May 21, 2026
7822d38
[nvbugs/6186880][fix] Drop kv_flat.contiguous() workaround per reviewer
tensorrt-cicd May 21, 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
7 changes: 6 additions & 1 deletion tensorrt_llm/_torch/modules/fused_moe/communication/deep_ep.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -190,7 +190,12 @@ def dispatch(
if token_final_scales is not None and token_final_scales.dtype != torch.float32:
token_final_scales = token_final_scales.to(torch.float32)

if not self.supports_post_quant_dispatch():
# Even when the comm strategy declares post-quant dispatch support, the
# caller may still pass `hidden_states_sf=None` if the local MoE module
# is excluded from quantization (e.g. an MTP layer with a non-quantized
# MoE in a partially-quantized checkpoint). In that case the data is
# actually unquantized, so dispatch the plain hidden_states.
if not self.supports_post_quant_dispatch() or hidden_states_sf is None:
# Pre-quant dispatch (unquantized data)
(
hidden_states,
Expand Down
4 changes: 0 additions & 4 deletions tests/integration/test_lists/waives.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,10 +32,6 @@ accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[disable_s
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_fp8_blockscale[latency_default] SKIP (https://nvbugs/6185196)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline] SKIP (https://nvbugs/6185196)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[baseline_mtp1] SKIP (https://nvbugs/6185196)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[fp4_indexer_dsl_mtp0] SKIP (https://nvbugs/6186880)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[fp4_indexer_dsl_mtp1] SKIP (https://nvbugs/6186880)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[fp4_indexer_dsl_mtp2] SKIP (https://nvbugs/6186880)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus[fp4_indexer_dsl_mtp3] SKIP (https://nvbugs/6186880)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[baseline] SKIP (https://nvbugs/6185196)
accuracy/test_llm_api_pytorch.py::TestDeepSeekV32::test_nvfp4_multi_gpus_piecewise_cuda_graph[mtp3_fp8kv_chunked] SKIP (https://nvbugs/5989920)
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] SKIP (https://nvbugs/6084720)
Expand Down
Loading