biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bug] Fix TRTLLM Block FP8 MoE Monolithic (#36296)

Browse Source 
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
Wei Zhao
2026-03-10 22:04:47 -04:00
committed by GitHub
parent 81939e7733
commit 84e436ed1c
1 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py
View File

@@ -176,9 +176,6 @@ class TrtLlmFp8Experts(mk.FusedMoEExpertsMonolithic):
assert not apply_router_weight_on_input
assert activation == MoEActivation.SILU
if e_score_correction_bias is not None:
e_score_correction_bias = e_score_correction_bias.to(hidden_states.dtype)
if self.routing_method_type == RoutingMethodType.DeepSeekV3:
router_logits = router_logits.to(torch.float32)