#!/bin/bash
set -e

echo "==================================="
echo "PaddleOCR-VL Server"
echo "==================================="

# Configuration
MODEL_NAME="${MODEL_NAME:-PaddlePaddle/PaddleOCR-VL}"
HOST="${HOST:-0.0.0.0}"
PORT="${PORT:-8000}"
MAX_BATCHED_TOKENS="${MAX_BATCHED_TOKENS:-16384}"
GPU_MEMORY_UTILIZATION="${GPU_MEMORY_UTILIZATION:-0.9}"

echo "Model: ${MODEL_NAME}"
echo "Host: ${HOST}"
echo "Port: ${PORT}"
echo "Max batched tokens: ${MAX_BATCHED_TOKENS}"
echo "GPU memory utilization: ${GPU_MEMORY_UTILIZATION}"
echo ""

# Check GPU availability
if command -v nvidia-smi &> /dev/null; then
    echo "GPU Information:"
    nvidia-smi --query-gpu=name,memory.total,memory.free --format=csv
    echo ""
else
    echo "WARNING: nvidia-smi not found. GPU may not be available."
fi

echo "Starting vLLM server..."
echo "==================================="

# Start vLLM server with PaddleOCR-VL
exec vllm serve "${MODEL_NAME}" \
    --trust-remote-code \
    --host "${HOST}" \
    --port "${PORT}" \
    --max-num-batched-tokens "${MAX_BATCHED_TOKENS}" \
    --gpu-memory-utilization "${GPU_MEMORY_UTILIZATION}" \
    --no-enable-prefix-caching \
    --mm-processor-cache-gb 0 \
    --served-model-name "paddleocr-vl"