Code hangs during inference deployment with vLLM

I encountered an issue where the code hangs during inference deployment with vLLM. How can I resolve this problem?

(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:46] Route: /v1/chat/completions/batch, Methods: POST
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/audio/speech/stream, Endpoint: streaming_speech
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/video/chat/stream, Endpoint: streaming_video_chat
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/realtime/video, Endpoint: streaming_video_output
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/realtime, Endpoint: realtime_websocket
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/realtime/robot/openpi, Endpoint: realtime_robot_openpi
(APIServer pid=2311773) INFO 08-10 12:10:18 [launcher.py:57] Route: /v1/duplex, Endpoint: duplex_websocket
(APIServer pid=2311773) INFO: Started server process [2311773]
(APIServer pid=2311773) INFO: Waiting for application startup.
(APIServer pid=2311773) INFO: Application startup complete.

Start with the vLLM deployment entry point and the launcher.py routes shown in the logs, then reproduce the hang using the reported startup output. Identify where startup stops after “Application startup complete” and verify that inference requests complete without hanging.