main函数
()
| 912 | |
| 913 | |
| 914 | def main(): |
| 915 | """main函数""" |
| 916 | if args.local_data_parallel_id == 0: |
| 917 | if not load_engine(): |
| 918 | return |
| 919 | else: |
| 920 | if not load_data_service(): |
| 921 | return |
| 922 | api_server_logger.info("FastDeploy LLM engine initialized!\n") |
| 923 | if args.metrics_port is not None and args.metrics_port != args.port: |
| 924 | launch_metrics_server() |
| 925 | console_logger.info(f"Launching metrics service at http://{args.host}:{args.metrics_port}/metrics") |
| 926 | else: |
| 927 | console_logger.info(f"Launching metrics service at http://{args.host}:{args.port}/metrics") |
| 928 | console_logger.info(f"Launching chat completion service at http://{args.host}:{args.port}/v1/chat/completions") |
| 929 | console_logger.info(f"Launching completion service at http://{args.host}:{args.port}/v1/completions") |
| 930 | |
| 931 | launch_worker_monitor() |
| 932 | launch_controller_server() |
| 933 | launch_api_server() |
| 934 | |
| 935 | |
| 936 | if __name__ == "__main__": |
no test coverage detected