From 2bdbbc80804b2199cbf39b75e78cf7269d0383a4 Mon Sep 17 00:00:00 2001 From: danisereb Date: Wed, 16 Sep 2026 22:31:43 +0300 Subject: [PATCH] [BugFix] Fix is_supported of cutlass FP8 linear (selected and fails on A100) (#55884) Signed-off-by: Daniel Serebrenik --- vllm/model_executor/kernels/linear/scaled_mm/cutlass.py | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/kernels/linear/scaled_mm/cutlass.py b/vllm/model_executor/kernels/linear/scaled_mm/cutlass.py index 9f69ab0c73..da8c69eed0 100644 --- a/vllm/model_executor/kernels/linear/scaled_mm/cutlass.py +++ b/vllm/model_executor/kernels/linear/scaled_mm/cutlass.py @@ -17,6 +17,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import ( from vllm.model_executor.layers.quantization.utils.w8a8_utils import ( CUTLASS_BLOCK_FP8_SUPPORTED, convert_to_channelwise, + cutlass_fp8_supported, ) from vllm.model_executor.utils import set_weight_attrs from vllm.platforms import current_platform @@ -164,8 +165,8 @@ class CutlassFP8ScaledMMLinearKernel(FP8ScaledMMLinearKernel): def is_supported( cls, compute_capability: int | None = None ) -> tuple[bool, str | None]: - if not current_platform.is_cuda(): - return False, "requires CUDA." + if not cutlass_fp8_supported(): + return False, "CUTLASS FP8 kernels not available" return True, None @classmethod