Deprecate --disable-flashinfer and --disable-flashinfer-sampling (#2065)

2024-11-17 16:20:58 -08:00
parent 38625e2139
commit 11f881d173
3 changed files with 25 additions and 28 deletions
--- a/python/sglang/srt/utils.py
+++ b/python/sglang/srt/utils.py
@@ -71,6 +71,8 @@ def is_flashinfer_available():
    Check whether flashinfer is available.
    As of Oct. 6, 2024, it is only available on NVIDIA GPUs.
    """
+    if os.environ.get("SGLANG_IS_FLASHINFER_AVAILABLE", "true") == "false":
+        return False
    return torch.cuda.is_available() and not is_hip()