#!/bin/bash
# 多worker部署脚本 - 在单台机器上启动API服务
# 使用方法: ./deploy_multi_worker.sh [num_workers] [port]

set -e

# 配置参数
NUM_WORKERS=${1:-2}  # 默认2个worker
PORT=${2:-8080}      # 默认端口8080
GPU_ID=${3:-0}       # 默认GPU 0

DECODER_CONFIG_NAME="modded_dac_vq"
LLAMA_CHECKPOINT="/root/fish-checkpoints/s2-pro"
DECODER_CHECKPOINT="/root/fish-checkpoints/s2-pro/codec.pth"

# 设置环境变量
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
export CUDA_VISIBLE_DEVICES=${GPU_ID}

echo "========================================="
echo "Fish-Speech Multi-Worker Deployment"
echo "========================================="
echo "Workers: ${NUM_WORKERS}"
echo "Port: ${PORT}"
echo "GPU: ${GPU_ID}"
echo "========================================="

# 启动API服务
python tools/api_server.py \
  --listen 0.0.0.0:${PORT} \
  --llama-checkpoint-path ${LLAMA_CHECKPOINT} \
  --decoder-checkpoint-path ${DECODER_CHECKPOINT} \
  --decoder-config-name ${DECODER_CONFIG_NAME} \
  --half \
  --workers 1 \
  --num-workers ${NUM_WORKERS}