Tiny allow DeepGEMM on cu12.9 (#9858)

2025-09-01 14:29:56 +08:00
parent 7de2ce45b2
commit 065e523d7b
1 changed files with 0 additions and 3 deletions
--- a/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
+++ b/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
@@ -11,9 +11,6 @@ def _compute_enable_deep_gemm():
    sm_version = get_device_sm()
    if sm_version < 90:
        return False
-    # TODO fix deepgemm cu129 fp8 issue
-    if torch.version.cuda == "12.9":
-        return False

    try:
        import deep_gemm