From 6da1310f91d7ecb0b40a252880e1d5c0d163a6e4 Mon Sep 17 00:00:00 2001 From: Wei Zhao <51183510+wzhao18@users.noreply.github.com> Date: Tue, 10 Mar 2026 22:04:47 -0400 Subject: [PATCH] [Bug] Fix TRTLLM Block FP8 MoE Monolithic (#36296) Signed-off-by: wzhao18 Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com> (cherry picked from commit 84e436ed1c94b1b94f809927b5d6bff45f7af919) --- vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py b/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py index 64b772505..1ed76f892 100644 --- a/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py +++ b/vllm/model_executor/layers/fused_moe/experts/trtllm_fp8_moe.py @@ -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)