dashboard-nanobot/download_model.sh

75 lines
2.3 KiB
Bash
Raw Normal View History

#!/usr/bin/env bash
# ==========================================
# Whisper ggml 模型下载脚本
# 下载的模型将存放在 data/model 目录下
# ==========================================
# 确保在项目根目录运行
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
TARGET_DIR="${SCRIPT_DIR}/data/model"
# 创建目标目录
mkdir -p "$TARGET_DIR"
# Hugging Face 的 whisper.cpp 模型仓库地址
# 使用 hf-mirror 镜像加速国内下载
BASE_URL="https://hf-mirror.com/ggerganov/whisper.cpp/resolve/main"
echo "=========================================="
echo "请选择要下载的 Whisper 语音识别模型:"
echo "1) tiny (最小,速度最快,约 40MB)"
echo "2) base (基础,速度快,约 74MB)"
echo "3) small (当前默认,平衡,约 240MB)"
echo "4) medium (推荐,效果更好,约 770MB)"
echo "5) large-v3 (最大,效果最好,约 1.5GB)"
echo "=========================================="
read -p "请输入对应的数字 [1-5默认 4]: " choice
MODEL_NAME=""
case $choice in
1) MODEL_NAME="ggml-tiny-q8_0.bin" ;;
2) MODEL_NAME="ggml-base-q8_0.bin" ;;
3) MODEL_NAME="ggml-small-q8_0.bin" ;;
4|"") MODEL_NAME="ggml-medium-q8_0.bin" ;;
5) MODEL_NAME="ggml-large-v3-q8_0.bin" ;;
*)
echo "无效的选择,使用默认选项 4 (medium)"
MODEL_NAME="ggml-medium-q8_0.bin"
;;
esac
echo ""
echo "准备下载模型: $MODEL_NAME"
echo "保存路径: $TARGET_DIR/$MODEL_NAME"
echo ""
# 下载文件
DOWNLOAD_URL="${BASE_URL}/${MODEL_NAME}"
if command -v wget >/dev/null 2>&1; then
echo "使用 wget 下载中..."
wget -c --show-progress "$DOWNLOAD_URL" -O "$TARGET_DIR/$MODEL_NAME"
elif command -v curl >/dev/null 2>&1; then
echo "使用 curl 下载中..."
curl -L -C - -o "$TARGET_DIR/$MODEL_NAME" "$DOWNLOAD_URL"
else
echo "错误:找不到 wget 或 curl 命令,无法下载文件。"
exit 1
fi
if [ $? -eq 0 ]; then
echo ""
echo "✅ 下载成功!"
echo "模型文件已保存到: $TARGET_DIR/$MODEL_NAME"
echo ""
echo "请记得更新 .env.prod 或 backend/.env 文件中的配置:"
echo "STT_MODEL=$MODEL_NAME"
echo "修改后重启后端容器即可生效。"
else
echo ""
echo "❌ 下载失败,请检查网络连接或稍后重试。"
exit 1
fi