File tree 2 files changed +12
-4
lines changed
offline_inference/qwen2_5_omni
online_serving/disaggregated_serving
2 files changed +12
-4
lines changed Original file line number Diff line number Diff line change @@ -140,7 +140,7 @@ def main(args):
140
140
print (generated_text )
141
141
142
142
143
- if __name__ == "__main__" :
143
+ def parse_args () :
144
144
parser = FlexibleArgumentParser (
145
145
description = 'Demo on using vLLM for offline inference with '
146
146
'audio language models' )
@@ -155,5 +155,9 @@ def main(args):
155
155
default = None ,
156
156
help = "Set the seed when initializing `vllm.LLM`." )
157
157
158
- args = parser .parse_args ()
158
+ return parser .parse_args ()
159
+
160
+
161
+ if __name__ == "__main__" :
162
+ args = parse_args ()
159
163
main (args )
Original file line number Diff line number Diff line change @@ -414,7 +414,7 @@ def run_server(self):
414
414
server .run ()
415
415
416
416
417
- if __name__ == "__main__" :
417
+ def parse_args () :
418
418
# Todo: allow more config
419
419
parser = argparse .ArgumentParser ("vLLM disaggregated proxy server." )
420
420
parser .add_argument ("--model" ,
@@ -445,6 +445,10 @@ def run_server(self):
445
445
default = 8000 ,
446
446
help = "Server port number" ,
447
447
)
448
- args = parser .parse_args ()
448
+ return parser .parse_args ()
449
+
450
+
451
+ if __name__ == "__main__" :
452
+ args = parse_args ()
449
453
proxy_server = ProxyServer (args = args )
450
454
proxy_server .run_server ()
You can’t perform that action at this time.
0 commit comments