From 39e8b49378090fed8df0a035cde0f47bb574a2de Mon Sep 17 00:00:00 2001
From: Michael Goin <mgoin64@gmail.com>
Date: Wed, 28 Jan 2026 21:40:59 -0500
Subject: [PATCH] [Bugfix] Register fp8 cutlass_group_gemm as supported for
 only SM90+SM100 (#33285)

Signed-off-by: mgoin <mgoin64@gmail.com>
(cherry picked from commit 1bd47d6e5a57000e404af60838eadb86b6f457c0)
---
 vllm/_custom_ops.py | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/vllm/_custom_ops.py b/vllm/_custom_ops.py
index 20f399d7f..92d21b742 100644
--- a/vllm/_custom_ops.py
+++ b/vllm/_custom_ops.py
@@ -900,6 +900,8 @@ def cutlass_sparse_scaled_mm_supported(cuda_device_capability: int) -> bool:
 
 
 def cutlass_group_gemm_supported(cuda_device_capability: int) -> bool:
+    if cuda_device_capability < 90 or cuda_device_capability >= 110:
+        return False
     try:
         return torch.ops._C.cutlass_group_gemm_supported(cuda_device_capability)
     except AttributeError: