From 0d4d164488cb29bdc9fbfe3cd943da4b854ae19c Mon Sep 17 00:00:00 2001 From: Gabriel Wu <13583761+lucifer1004@users.noreply.github.com> Date: Wed, 24 Jun 2026 03:43:36 +0800 Subject: [PATCH] [Bugfix] Allow flashinfer_cutlass as a clamped NVFP4 MoE backend (#46492) Signed-off-by: Zihua Wu <13583761+lucifer1004@users.noreply.github.com> Signed-off-by: Michael Goin Co-authored-by: Claude Opus 4.8 (1M context) Co-authored-by: Michael Goin --- vllm/model_executor/layers/fused_moe/oracle/nvfp4.py | 1 + 1 file changed, 1 insertion(+) diff --git a/vllm/model_executor/layers/fused_moe/oracle/nvfp4.py b/vllm/model_executor/layers/fused_moe/oracle/nvfp4.py index 8603ca85b8..408c69fea0 100644 --- a/vllm/model_executor/layers/fused_moe/oracle/nvfp4.py +++ b/vllm/model_executor/layers/fused_moe/oracle/nvfp4.py @@ -174,6 +174,7 @@ def select_nvfp4_moe_backend( NVFP4_BACKENDS_WITH_CLAMP = { NvFp4MoeBackend.FLASHINFER_TRTLLM, + NvFp4MoeBackend.FLASHINFER_CUTLASS, NvFp4MoeBackend.MARLIN, }