A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 19:45 UTC
[Build] Add nightly wheel fallback when latest commit wheel unavailable (#14358)
Signed-off-by: Isotr0py <2037008807@qq.com>
Isotr0py committed
Mar 7, 2025 at 14:10 UTC
63137cd9226c8ee335b8c0dfbb002671d6a917e1
1 file changed
+16
setup.py
+16
@@ -69,6 +69,18 @@ def is_ninja_available() -> bool:
69
return which("ninja") is not None
70
71
72
+def is_url_available(url: str) -> bool:
73
+ from urllib.request import urlopen
74
+
75
+ status = None
76
+ try:
77
+ with urlopen(url) as f:
78
+ status = f.status
79
+ except Exception:
80
+ return False
81
+ return status == 200
82
+
83
+
84
class CMakeExtension(Extension):
85
86
def __init__(self, name: str, cmake_lists_dir: str = '.', **kwa) -> None:
@@ -320,6 +332,10 @@ class repackage_wheel(build_ext):
332
if wheel_location is None:
333
base_commit = self.get_base_commit_in_main_branch()
334
wheel_location = f"https://wheels.vllm.ai/{base_commit}/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl"
335
+ # Fallback to nightly wheel if latest commit wheel is unavailable,
336
+ # in this rare case, the nightly release CI hasn't finished on main.
337
+ if not is_url_available(wheel_location):
338
+ wheel_location = "https://wheels.vllm.ai/nightly/vllm-1.0.0.dev-cp38-abi3-manylinux1_x86_64.whl"
339
340
import zipfile
341