local-llm-server/other/local-llm-server.service

[Unit]
Description=Local LLM Proxy Server
Wants=basic.target
After=basic.target network.target local-llm-daemon.service
Requires=local-llm-daemon.service

[Service]
User=server
WorkingDirectory=/srv/server/local-llm-server

# Sometimes the old processes aren't terminated when the service is restarted.
ExecStartPre=/usr/bin/pkill -9 -f "/srv/server/local-llm-server/venv/bin/python3 /srv/server/local-llm-server/venv/bin/gunicorn"

# TODO: make sure gunicorn logs to stdout and logging also goes to stdout

# Need a lot of workers since we have long-running requests. This takes about 3.5G memory.
ExecStart=/srv/server/local-llm-server/venv/bin/gunicorn -c other/gunicorn_conf.py --workers 20 --bind 0.0.0.0:5000 server:app --timeout 60 --worker-class gevent --access-logfile '-' --error-logfile '-'

Restart=always
RestartSec=2
SyslogIdentifier=local-llm-server

[Install]
WantedBy=multi-user.target