From c807cd7c751d3709f90d95fc32b0757e2af0d3bb Mon Sep 17 00:00:00 2001
From: Yineng Zhang <me@zhyncs.com>
Date: Sun, 24 Aug 2025 01:05:00 -0700
Subject: [PATCH] chore: update configurer (#9557)

---
 .../layers/quantization/deep_gemm_wrapper/configurer.py   | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py b/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
index 10fb2e7ba..cb4c2edb1 100644
--- a/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
+++ b/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
@@ -1,5 +1,7 @@
 import logging
 
+import torch
+
 from sglang.srt.utils import get_bool_env_var, get_device_sm
 
 logger = logging.getLogger(__name__)
@@ -7,8 +9,10 @@ logger = logging.getLogger(__name__)
 
 def _compute_enable_deep_gemm():
     sm_version = get_device_sm()
-    # TODO fix blackwell fp8
-    if sm_version != 90:
+    if sm_version < 90:
+        return False
+    # TODO fix deepgemm cu129 fp8 issue
+    if torch.version.cuda == "12.9":
         return False
 
     try: