add control for cutlass fp8 blockwise gemm (#3727)

1eb8eade · yizhang2077 · GitHub · 3c7bfd7e · 1eb8eade
Unverified Commit 1eb8eade authored Feb 20, 2025 by yizhang2077 Committed by GitHub Feb 20, 2025
Hide whitespace changes
Inline Side-by-side

Showing with 3 additions and 0 deletions

python/sglang/srt/layers/quantization/fp8_utils.py python/sglang/srt/layers/quantization/fp8_utils.py +3 -0

No files found.
--- a/python/sglang/srt/layers/quantization/fp8_utils.py
+++ b/python/sglang/srt/layers/quantization/fp8_utils.py
+import os
 from typing import List, Optional, Tuple

 import torch
@@ -40,6 +41,8 @@ def normalize_e4m3fn_to_e4m3fnuz(


 def cutlass_block_fp8_supported() -> bool:
+    if os.environ.get("SUPPORT_CUTLASS_BLOCK_FP8") is None:
+        return False
    if _is_cuda:
        major, minor = torch.cuda.get_device_capability()
        sm_version = major * 10 + minor