fix: use deepgemm only on hopper (#5310)

61172091 · Yineng Zhang · GitHub · 4f288113 · 61172091
Unverified Commit 61172091 authored Apr 11, 2025 by Yineng Zhang Committed by GitHub Apr 11, 2025
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

python/sglang/srt/layers/quantization/fp8_kernel.py python/sglang/srt/layers/quantization/fp8_kernel.py +1 -1

No files found.
--- a/python/sglang/srt/layers/quantization/fp8_kernel.py
+++ b/python/sglang/srt/layers/quantization/fp8_kernel.py
@@ -45,7 +45,7 @@ if _is_cuda:
    from sgl_kernel import sgl_per_token_group_quant_fp8, sgl_per_token_quant_fp8

    sm_version = get_device_sm()
-    if sm_version >= 90 and get_bool_env_var("SGL_ENABLE_JIT_DEEPGEMM", default="true"):
+    if sm_version == 90 and get_bool_env_var("SGL_ENABLE_JIT_DEEPGEMM", default="true"):
        _enable_jit_deepgemm = True