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, 09:45 UTC
[Misc] extract parser.parse_args() (#18323)
Signed-off-by: reidliu41 <reid201711@gmail.com> Co-authored-by: reidliu41 <reid201711@gmail.com>
Reid committed
May 19, 2025 at 12:06 UTC
27d09526006a8d437874628d6905df12f80a194f
2 files changed
+12
-4
examples/offline_inference/qwen2_5_omni/only_thinker.py
+6
-2
@@ -140,7 +140,7 @@ def main(args):
140
print(generated_text)
141
142
143
-if __name__ == "__main__":
143
+def parse_args():
144
parser = FlexibleArgumentParser(
145
description='Demo on using vLLM for offline inference with '
146
'audio language models')
@@ -155,5 +155,9 @@ if __name__ == "__main__":
155
default=None,
156
help="Set the seed when initializing `vllm.LLM`.")
157
158
- args = parser.parse_args()
158
+ return parser.parse_args()
159
+
160
+
161
+if __name__ == "__main__":
162
+ args = parse_args()
163
main(args)
examples/online_serving/disaggregated_serving/disagg_proxy_demo.py
+6
-2
@@ -414,7 +414,7 @@ class ProxyServer:
414
server.run()
415
416
417
-if __name__ == "__main__":
417
+def parse_args():
418
# Todo: allow more config
419
parser = argparse.ArgumentParser("vLLM disaggregated proxy server.")
420
parser.add_argument("--model",
@@ -445,6 +445,10 @@ if __name__ == "__main__":
445
default=8000,
446
help="Server port number",
447
)
448
- args = parser.parse_args()
448
+ return parser.parse_args()
449
+
450
+
451
+if __name__ == "__main__":
452
+ args = parse_args()
453
proxy_server = ProxyServer(args=args)
454
proxy_server.run_server()