24 lines
824 B
Plaintext
24 lines
824 B
Plaintext
[supervisord]
|
|
nodaemon=true
|
|
logfile=/dev/stdout
|
|
logfile_maxbytes=0
|
|
loglevel=info
|
|
|
|
[program:sglang]
|
|
command=python3 -m sglang.launch_server --host 0.0.0.0 --port 30000 --model-path /root/.cradle/external/llm/ --lora-paths q3=/root/.cradle/external/lora/q3 --lora-target-modules q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj --max-lora-rank 16 --served-model-name qwen3-32b --disable-radix-cache --tp 4 --api-key token-abc123 --enable-metrics --log-requests --log-requests-level 2
|
|
autostart=true
|
|
autorestart=true
|
|
stdout_logfile=/dev/stdout
|
|
stdout_logfile_maxbytes=0
|
|
stderr_logfile=/dev/stderr
|
|
stderr_logfile_maxbytes=0
|
|
|
|
[program:ui]
|
|
command=python3 /app/meta_ui.py --port 30001
|
|
autostart=true
|
|
autorestart=true
|
|
stdout_logfile=/dev/stdout
|
|
stdout_logfile_maxbytes=0
|
|
stderr_logfile=/dev/stderr
|
|
stderr_logfile_maxbytes=0
|