From 29026682cb29335dcea15d82c14ee18fff40f71d Mon Sep 17 00:00:00 2001 From: bnellnm <49004751+bnellnm@users.noreply.github.com> Date: Wed, 10 Jun 2026 13:16:25 -0400 Subject: [PATCH] [Bugfix] Fix nemotron accuracy drop introduced by #41184 (#45037) Signed-off-by: Bill Nell --- vllm/model_executor/layers/fused_moe/layer.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vllm/model_executor/layers/fused_moe/layer.py b/vllm/model_executor/layers/fused_moe/layer.py index 815e5c5fb7..15806ca4f8 100644 --- a/vllm/model_executor/layers/fused_moe/layer.py +++ b/vllm/model_executor/layers/fused_moe/layer.py @@ -349,7 +349,9 @@ def FusedMoE( topk_group=topk_group, custom_routing_function=custom_routing_function, scoring_func=scoring_func, - routed_scaling_factor=routed_scaling_factor, + routed_scaling_factor=routed_scaling_factor + if not apply_routed_scale_to_output + else 1.0, swiglu_limit=swiglu_limit, # TODO get from router? needs to be truncated? e_score_correction_bias=e_score_correction_bias,