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 10, 10:15 UTC
[CI] [ROCm] Fix `tests/entrypoints/test_grpc_server.py` on ROCm (#31970)
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
TJian committed
Jan 9, 2026 at 12:54 UTC
7a05d2dc65065da3d6fd54c67a670ab08ffaf600
5 files changed
+18
-4
requirements/common.txt
+1
-1
@@ -53,4 +53,4 @@ anthropic == 0.71.0
53
model-hosting-container-standards >= 0.1.10, < 1.0.0
54
mcp
55
grpcio>=1.76.0
56
-grpcio-reflection>=1.76.0
56
+grpcio-reflection>=1.76.0
\ No newline at end of file
requirements/rocm-test.txt
+1
-1
@@ -89,4 +89,4 @@ perceptron==0.1.4
89
# Required for the multi-modal models test
90
timm==1.0.17
91
# Required for plugins test
92
-albumentations==1.4.6
92
+albumentations==1.4.6
\ No newline at end of file
requirements/rocm.txt
+1
@@ -16,3 +16,4 @@ runai-model-streamer[s3,gcs]==0.15.3
16
conch-triton-kernels==1.2.1
17
timm>=1.0.17
18
fastsafetensors @ git+https://github.com/foundation-model-stack/fastsafetensors.git@d6f998a03432b2452f8de2bb5cefb5af9795d459
19
+grpcio-tools>=1.76.0
\ No newline at end of file
setup.py
+14
-1
@@ -19,6 +19,7 @@ from packaging.version import Version, parse
19
from setuptools import Extension, setup
20
from setuptools.command.build_ext import build_ext
21
from setuptools.command.build_py import build_py
22
+from setuptools.command.develop import develop
23
from setuptools_scm import get_version
24
from torch.utils.cpp_extension import CUDA_HOME, ROCM_HOME
25
@@ -147,6 +148,14 @@ class BuildPyAndGenerateGrpc(build_py):
148
super().run()
149
150
151
+class DevelopAndGenerateGrpc(develop):
152
+ """Develop mode that also generates gRPC stubs from proto files."""
153
+
154
+ def run(self):
155
+ compile_grpc_protos()
156
+ super().run()
157
+
158
+
159
class CMakeExtension(Extension):
160
def __init__(self, name: str, cmake_lists_dir: str = ".", **kwa) -> None:
161
super().__init__(name, sources=[], py_limited_api=not is_freethreaded(), **kwa)
@@ -950,13 +959,17 @@ if _no_device():
959
ext_modules = []
960
961
if not ext_modules:
953
- cmdclass = {"build_py": BuildPyAndGenerateGrpc}
962
+ cmdclass = {
963
+ "build_py": BuildPyAndGenerateGrpc,
964
+ "develop": DevelopAndGenerateGrpc,
965
+ }
966
else:
967
cmdclass = {
968
"build_ext": precompiled_build_ext
969
if envs.VLLM_USE_PRECOMPILED
970
else cmake_build_ext,
971
"build_py": BuildPyAndGenerateGrpc,
972
+ "develop": DevelopAndGenerateGrpc,
973
}
974
975
setup(
tests/entrypoints/test_grpc_server.py
+1
-1
@@ -29,7 +29,7 @@ def find_free_port() -> int:
29
return port
30
31
32
-async def wait_for_server(port: int, timeout: float = 30.0) -> bool:
32
+async def wait_for_server(port: int, timeout: float = 60.0) -> bool:
33
"""Wait for the gRPC server to be ready by trying health checks."""
34
start_time = time.time()
35
print("waiting for server to start...")