dots.ocr/zhch/vllm_daemon.sh

#!/bin/bash

LOGDIR="/home/ubuntu/zhch/logs"
mkdir -p $LOGDIR
PIDFILE="$LOGDIR/vllm.pid"
LOGFILE="$LOGDIR/vllm.log"

conda activate py312
hf_model_path="/home/ubuntu/zhch/dots.ocr/weights/DotsOCR"  # Path to your downloaded model weights
export PYTHONPATH=$(dirname "$hf_model_path"):$PYTHONPATH
# launch vllm server
model_name="DotsOCR"

start() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "vLLM is already running"
        return 1
    fi
    
    echo "Starting vLLM daemon..."
    nohup vllm serve $hf_model_path \
		--host 0.0.0.0 \
		--port 8101 \
		--gpu-memory-utilization 0.90 \
		--max-log-len 1000 \
		--trust-remote-code \
		--max-model-len 65536 \
		--max-num-batched-tokens 65536 \
		--uvicorn-log-level info \
		--limit-mm-per-prompt '{"image": 5}' \
		--chat-template-content-format string \
		--data-parallel-size 3 \
		--max-num-seqs 32 \
		--enable-prefix-caching \
		--served-model-name ${model_name} \
		> $LOGFILE 2>&1 &
    
    echo $! > $PIDFILE
    echo "vLLM started with PID: $(cat $PIDFILE)"
}

stop() {
    if [ ! -f $PIDFILE ]; then
        echo "vLLM is not running"
        return 1
    fi
    
    PID=$(cat $PIDFILE)
    echo "Stopping vLLM (PID: $PID)..."
    kill $PID
    rm -f $PIDFILE
    echo "vLLM stopped"
}

status() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "vLLM is running (PID: $(cat $PIDFILE))"
    else
        echo "vLLM is not running"
    fi
}

case "$1" in
    start)
        start
        ;;
    stop)
        stop
        ;;
    restart)
        stop
        sleep 2
        start
        ;;
    status)
        status
        ;;
    *)
        echo "Usage: $0 {start|stop|restart|status}"
        exit 1
        ;;
esac