ocr_platform/ocr_tools/daemons/glmocr_local_daemon.sh

#!/bin/bash
# filepath: ocr_platform/ocr_tools/daemons/glmocr_local_daemon.sh
# 对应: GLM-OCR 本地 llama-server 服务（macOS），使用 GGUF 格式模型
# 适用于 Mac M4 Pro 48G，使用 Metal GPU 加速
# 模型下载地址: https://huggingface.co/ggml-org/GLM-OCR-GGUF
# 模型下载地址: https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5-GGUF

# unset https_proxy http_proxy HF_ENDPOINT
# llama-server -hf ggml-org/GLM-OCR-GGUF:Q8_0
# mv ~/Library/Caches/llama.cpp/ggml-org_GLM-OCR-GGUF_GLM-OCR-Q8_0.gguf  ~/models/glmocr
# mv ~/Library/Caches/llama.cpp/ggml-org_GLM-OCR-GGUF_mmproj-GLM-OCR-Q8_0.gguf  ~/models/glmocr

# curl -X POST http://localhost:8101/v1/chat/completions -d @payload.json

LOGDIR="$HOME/workspace/logs"
mkdir -p $LOGDIR
PIDFILE="$LOGDIR/glmocr_llamaserver.pid"
LOGFILE="$LOGDIR/glmocr_llamaserver.log"

# 配置参数
CONDA_ENV="mineru2"
PORT="8101"
HOST="0.0.0.0"

# 本地 GGUF 模型路径
MODEL_PATH="$HOME/models/glmocr/ggml-org_GLM-OCR-GGUF_GLM-OCR-Q8_0.gguf"
MMPROJ_PATH="$HOME/models/glmocr/ggml-org_GLM-OCR-GGUF_mmproj-GLM-OCR-Q8_0.gguf"

# llama-server 参数
CONTEXT_SIZE="16384"         # 上下文长度（需 >= max_tokens，推荐 8192-16384）
GPU_LAYERS="99"              # Metal GPU 层数（99 表示全部）
THREADS="8"                  # CPU 线程数（M4 Pro 建议值）
BATCH_SIZE="512"             # 批处理大小
UBATCH_SIZE="128"            # 微批处理大小

# conda 环境激活
if [ -f "$HOME/anaconda3/etc/profile.d/conda.sh" ]; then
    source "$HOME/anaconda3/etc/profile.d/conda.sh"
    conda activate $CONDA_ENV
elif [ -f "$HOME/miniconda3/etc/profile.d/conda.sh" ]; then
    source "$HOME/miniconda3/etc/profile.d/conda.sh"
    conda activate $CONDA_ENV
elif [ -f "/opt/miniconda3/etc/profile.d/conda.sh" ]; then
    source /opt/miniconda3/etc/profile.d/conda.sh
    conda activate $CONDA_ENV
else
    echo "Warning: conda initialization file not found, trying direct path"
    export PATH="/opt/miniconda3/envs/$CONDA_ENV/bin:$PATH"
fi

start() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "GLM-OCR llama-server 已在运行"
        return 1
    fi

    echo "启动 GLM-OCR llama-server 守护进程..."
    echo "Host: $HOST, Port: $PORT"
    echo "主模型: $MODEL_PATH"
    echo "多模态投影器: $MMPROJ_PATH"
    echo "上下文长度: $CONTEXT_SIZE"
    echo "GPU 层数: $GPU_LAYERS (Metal)"
    echo "线程数: $THREADS"

    # 检查模型文件是否存在
    if [ ! -f "$MODEL_PATH" ]; then
        echo "❌ 主模型文件不存在: $MODEL_PATH"
        echo "请确认模型已下载到 llama.cpp 缓存目录"
        return 1
    fi

    if [ ! -f "$MMPROJ_PATH" ]; then
        echo "❌ 多模态投影器文件不存在: $MMPROJ_PATH"
        echo "请确认 mmproj 文件已下载"
        return 1
    fi

    # 检查 llama-server 命令
    if ! command -v llama-server >/dev/null 2>&1; then
        echo "❌ llama-server 未找到"
        echo "请安装: brew install llama.cpp"
        return 1
    fi

    echo "🔧 使用 llama-server: $(which llama-server)"
    echo "🔧 llama.cpp 版本: $(llama-server --version 2>&1 | head -1 || echo 'Unknown')"

    echo "💻 系统信息:"
    echo "  架构: $(uname -m)"
    echo "  系统: $(uname -s)"
    echo "  内存: $(sysctl -n hw.memsize | awk '{printf "%.1f GB", $1/1024/1024/1024}')"

    # 启动 llama-server
    # --log-disable \
    nohup llama-server \
        -m "$MODEL_PATH" \
        --mmproj "$MMPROJ_PATH" \
        --host $HOST \
        --port $PORT \
        --media-path $HOME/workspace \
        -c $CONTEXT_SIZE \
        -ngl $GPU_LAYERS \
        -t $THREADS \
        -b $BATCH_SIZE \
        -ub $UBATCH_SIZE \
        --temp 0 \
        > $LOGFILE 2>&1 &

    echo $! > $PIDFILE
    echo "✅ GLM-OCR llama-server 已启动，PID: $(cat $PIDFILE)"
    echo "📋 日志文件: $LOGFILE"
    echo "🌐 服务 URL: http://$HOST:$PORT"
    echo "📖 OpenAI 兼容 API: http://localhost:$PORT/v1 (chat/completions, models)"
    echo ""
    echo "等待服务启动..."
    sleep 5
    status
}

stop() {
    if [ ! -f $PIDFILE ]; then
        echo "GLM-OCR llama-server 未在运行"
        return 1
    fi

    PID=$(cat $PIDFILE)
    echo "停止 GLM-OCR llama-server (PID: $PID)..."

    kill $PID

    for i in {1..30}; do
        if ! kill -0 $PID 2>/dev/null; then
            break
        fi
        echo "等待进程停止... ($i/30)"
        sleep 1
    done

    if kill -0 $PID 2>/dev/null; then
        echo "强制终止进程..."
        kill -9 $PID
    fi

    rm -f $PIDFILE
    echo "✅ GLM-OCR llama-server 已停止"
}

status() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        PID=$(cat $PIDFILE)
        echo "✅ GLM-OCR llama-server 正在运行 (PID: $PID)"
        echo "🌐 服务 URL: http://$HOST:$PORT"
        echo "📋 日志文件: $LOGFILE"

        # 检查端口监听状态
        if lsof -nP -iTCP:$PORT -sTCP:LISTEN >/dev/null 2>&1; then
            echo "🔗 端口 $PORT 正在监听"
        else
            echo "⚠️  端口 $PORT 未在监听（服务可能正在启动）"
        fi

        # 检查 API 响应
        if command -v curl >/dev/null 2>&1; then
            if curl -s --connect-timeout 2 http://127.0.0.1:$PORT/v1/models > /dev/null 2>&1; then
                echo "🎯 API 响应正常"
            else
                echo "⚠️  API 无响应（服务可能正在启动）"
            fi
        fi

        # 显示进程内存使用
        if command -v ps >/dev/null 2>&1; then
            MEM=$(ps -o rss= -p $PID 2>/dev/null | awk '{printf "%.2f GB", $1/1024/1024}')
            if [ -n "$MEM" ]; then
                echo "💾 内存使用: $MEM"
            fi
        fi

        if [ -f $LOGFILE ]; then
            echo "📄 最近日志（最后 3 行）:"
            tail -3 $LOGFILE | sed 's/^/  /'
        fi
    else
        echo "❌ GLM-OCR llama-server 未在运行"
        if [ -f $PIDFILE ]; then
            echo "删除过期的 PID 文件..."
            rm -f $PIDFILE
        fi
    fi
}

logs() {
    if [ -f $LOGFILE ]; then
        echo "📄 GLM-OCR llama-server 日志:"
        echo "====================="
        tail -f $LOGFILE
    else
        echo "❌ 日志文件不存在: $LOGFILE"
    fi
}

config() {
    echo "📋 当前配置:"
    echo "  Conda 环境: $CONDA_ENV"
    echo "  Host: $HOST"
    echo "  Port: $PORT"
    echo "  主模型路径: $MODEL_PATH"
    echo "  多模态投影器: $MMPROJ_PATH"
    echo "  上下文长度: $CONTEXT_SIZE"
    echo "  GPU 层数: $GPU_LAYERS"
    echo "  线程数: $THREADS"
    echo "  批处理大小: $BATCH_SIZE"
    echo "  微批处理大小: $UBATCH_SIZE"
    echo "  PID 文件: $PIDFILE"
    echo "  日志文件: $LOGFILE"

    echo ""
    echo "📦 模型文件检查:"
    if [ -f "$MODEL_PATH" ]; then
        SIZE=$(du -h "$MODEL_PATH" | cut -f1)
        echo "  ✅ 主模型存在 ($SIZE)"
    else
        echo "  ❌ 主模型不存在"
    fi

    if [ -f "$MMPROJ_PATH" ]; then
        SIZE=$(du -h "$MMPROJ_PATH" | cut -f1)
        echo "  ✅ 多模态投影器存在 ($SIZE)"
    else
        echo "  ❌ 多模态投影器不存在"
    fi

    echo ""
    echo "🔧 环境检查:"
    echo "  llama-server: $(which llama-server 2>/dev/null || echo '未安装')"
    if command -v llama-server >/dev/null 2>&1; then
        LLAMA_VERSION=$(llama-server --version 2>&1 | head -1 || echo 'Unknown')
        echo "  版本: $LLAMA_VERSION"
    fi
    echo "  Conda: $(which conda 2>/dev/null || echo '未找到')"
    echo "  当前 Python: $(which python 2>/dev/null || echo '未找到')"

    echo ""
    echo "💻 系统信息:"
    echo "  架构: $(uname -m)"
    echo "  系统版本: $(sw_vers -productVersion 2>/dev/null || echo 'Unknown')"
    echo "  总内存: $(sysctl -n hw.memsize 2>/dev/null | awk '{printf "%.1f GB", $1/1024/1024/1024}' || echo 'Unknown')"
    echo "  CPU 核心: $(sysctl -n hw.ncpu 2>/dev/null || echo 'Unknown')"
}

test_api() {
    echo "🧪 测试 GLM-OCR llama-server API..."

    if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "❌ GLM-OCR llama-server 服务未在运行"
        return 1
    fi

    if ! command -v curl >/dev/null 2>&1; then
        echo "❌ curl 命令未找到"
        return 1
    fi

    echo "📡 测试 /v1/models 端点..."
    response=$(curl -s --connect-timeout 10 http://127.0.0.1:$PORT/v1/models)
    if [ $? -eq 0 ]; then
        echo "✅ Models 端点可访问"
        echo "$response" | python -m json.tool 2>/dev/null || echo "$response"
    else
        echo "❌ Models 端点不可访问"
    fi

    echo ""
    echo "📡 测试 /health 端点..."
    health=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/health)
    if [ $? -eq 0 ]; then
        echo "✅ Health 端点: $health"
    else
        echo "⚠️  Health 端点不可访问"
    fi
}

test_client() {
    echo "🧪 测试 GLM-OCR 与 llama-server 集成..."

    if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "❌ GLM-OCR llama-server 服务未在运行，请先启动: $0 start"
        return 1
    fi

    CONFIG_FILE="/Users/zhch158/workspace/repository.git/ocr_platform/ocr_tools/universal_doc_parser/config/bank_statement_yusys_local.yaml"
    
    echo "📄 配置文件: $CONFIG_FILE"
    echo ""
    echo "确保配置文件中 vl_recognition.api_url 指向: http://localhost:$PORT/v1/chat/completions"
    echo ""
    echo "测试命令示例:"
    echo "  cd /Users/zhch158/workspace/repository.git/ocr_platform/ocr_tools/universal_doc_parser"
    echo "  conda activate mineru2"
    echo "  python parse.py --input /path/to/test/image.png --config $CONFIG_FILE --debug"
    echo ""
    echo "或者使用 curl 直接测试 API:"
    echo "  curl -X POST http://localhost:$PORT/v1/chat/completions \\"
    echo "    -H 'Content-Type: application/json' \\"
    echo "    -d '{"
    echo "      \"model\": \"glm-ocr\","
    echo "      \"messages\": ["
    echo "        {"
    echo "          \"role\": \"user\","
    echo "          \"content\": ["
    echo "            {\"type\": \"text\", \"text\": \"Table Recognition:\"},"
    echo "            {\"type\": \"image_url\", \"image_url\": {\"url\": \"file:///path/to/image.png\"}}"
    echo "          ]"
    echo "        }"
    echo "      ],"
    echo "      \"max_tokens\": 4096"
    echo "    }'"
}

usage() {
    echo "GLM-OCR llama-server 服务守护进程（macOS）"
    echo "==========================================="
    echo "用法: $0 {start|stop|restart|status|logs|config|test|test-client}"
    echo ""
    echo "命令:"
    echo "  start       - 启动 GLM-OCR llama-server 服务"
    echo "  stop        - 停止 GLM-OCR llama-server 服务"
    echo "  restart     - 重启 GLM-OCR llama-server 服务"
    echo "  status      - 显示服务状态和资源使用"
    echo "  logs        - 显示服务日志（跟踪模式）"
    echo "  config      - 显示当前配置"
    echo "  test        - 测试 /v1/models API 端点"
    echo "  test-client - 显示如何测试与配置文件集成"
    echo ""
    echo "配置（编辑脚本修改）:"
    echo "  Host: $HOST"
    echo "  Port: $PORT"
    echo "  主模型: $MODEL_PATH"
    echo "  多模态投影器: $MMPROJ_PATH"
    echo "  上下文长度: $CONTEXT_SIZE"
    echo "  GPU 层数: $GPU_LAYERS (Metal)"
    echo ""
    echo "示例:"
    echo "  ./glmocr_local_daemon.sh start"
    echo "  ./glmocr_local_daemon.sh status"
    echo "  ./glmocr_local_daemon.sh logs"
    echo "  ./glmocr_local_daemon.sh test"
    echo ""
    echo "前置要求:"
    echo "  1. 安装 llama.cpp: brew install llama.cpp"
    echo "  2. 模型文件位于: ~/Library/Caches/llama.cpp/"
    echo "  3. conda 环境 mineru2 已配置"
}

case "$1" in
    start)
        start
        ;;
    stop)
        stop
        ;;
    restart)
        stop
        sleep 3
        start
        ;;
    status)
        status
        ;;
    logs)
        logs
        ;;
    config)
        config
        ;;
    test)
        test_api
        ;;
    test-client)
        test_client
        ;;
    *)
        usage
        exit 1
        ;;
esac