add an env var for path to pre-downloaded flashinfer cubin files (vllm-project#22675)

Signed-off-by: Xiao Yu <[email protected]>

Author: 842974287

vllm/envs.py

@@ -158,6 +158,7 @@
     VLLM_ALLOW_CHUNKED_LOCAL_ATTN_WITH_HYBRID_KV_CACHE: bool = False
     VLLM_ENABLE_RESPONSES_API_STORE: bool = False
     VLLM_USE_TRTLLM_ATTENTION: Optional[str] = None
+    VLLM_HAS_FLASHINFER_CUBIN: bool = False
     VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8: bool = False
     VLLM_USE_FLASHINFER_MOE_MXFP4_BF16: bool = False
     VLLM_TUNED_CONFIG_FOLDER: Optional[str] = None
@@ -1105,6 +1106,11 @@ def get_vllm_port() -> Optional[int]:
     "VLLM_USE_TRTLLM_ATTENTION":
     lambda: os.getenv("VLLM_USE_TRTLLM_ATTENTION", None),
 
+    # If set, it means we pre-downloaded cubin files and flashinfer will
+    # read the cubin files directly.
+    "VLLM_HAS_FLASHINFER_CUBIN":
+    lambda: os.getenv("VLLM_HAS_FLASHINFER_CUBIN", False),
+
     # If set to 1, force the use of TRTLLM FP4 GEMM backend in flashinfer.
     # Otherwise, uses the first available of: flashinfer cutlass GEMM,
     # vllm cutlass GEMM, marlin GEMM.

vllm/utils/flashinfer.py

@@ -132,6 +132,11 @@ def has_nvidia_artifactory() -> bool:
     This checks connectivity to the kernel inference library artifactory
     which is required for downloading certain cubin kernels like TRTLLM FHMA.
     """
+    # Since FLASHINFER_CUBIN_DIR defines the pre-downloaded cubins path, when
+    # it's true, we could assume the cubins are available.
+    if envs.VLLM_HAS_FLASHINFER_CUBIN:
+        return True
+
     try:
         # Use a short timeout to avoid blocking for too long
         response = requests.get(FLASHINFER_CUBINS_REPOSITORY, timeout=5)