From e346e2d056a66bb84287e4fea049bde9a37bd72b Mon Sep 17 00:00:00 2001 From: Michael Goin Date: Tue, 3 Feb 2026 05:37:15 -0500 Subject: [PATCH] [Bugfix] Disable RoutingMethodType.[Renormalize,RenormalizeNaive] TRTLLM per-tensor FP8 MoE (#33620) Signed-off-by: mgoin --- .../layers/fused_moe/flashinfer_trtllm_moe.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py b/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py index a066535c5..43e02d510 100644 --- a/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py +++ b/vllm/model_executor/layers/fused_moe/flashinfer_trtllm_moe.py @@ -72,8 +72,10 @@ def _supports_routing_method( # NOTE(dbari): as above, potentially allow others here. return routing_method in [ RoutingMethodType.Llama4, - RoutingMethodType.Renormalize, - RoutingMethodType.RenormalizeNaive, + # NOTE(mgoin): Disabled to investigate accuracy issues. + # See https://github.com/vllm-project/vllm/issues/33532 + # RoutingMethodType.Renormalize, + # RoutingMethodType.RenormalizeNaive, ] else: raise ValueError("Unsupported quantization scheme.")