2024-06-08 02:06:52 -07:00
|
|
|
"""Launch the inference server."""
|
|
|
|
|
|
2024-11-28 13:24:47 +00:00
|
|
|
import os
|
2024-09-09 02:14:25 -07:00
|
|
|
import sys
|
2024-01-08 04:37:50 +00:00
|
|
|
|
2025-01-19 22:09:24 -08:00
|
|
|
from sglang.srt.entrypoints.http_server import launch_server
|
2024-09-09 02:14:25 -07:00
|
|
|
from sglang.srt.server_args import prepare_server_args
|
2024-11-28 00:22:39 -08:00
|
|
|
from sglang.srt.utils import kill_process_tree
|
2024-01-08 04:37:50 +00:00
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
2024-09-09 02:14:25 -07:00
|
|
|
server_args = prepare_server_args(sys.argv[1:])
|
2024-01-08 04:37:50 +00:00
|
|
|
|
2024-08-20 22:35:05 -07:00
|
|
|
try:
|
2024-09-09 04:14:11 -07:00
|
|
|
launch_server(server_args)
|
2024-08-20 22:35:05 -07:00
|
|
|
finally:
|
2024-11-28 00:22:39 -08:00
|
|
|
kill_process_tree(os.getpid(), include_parent=False)
|