paddle_local_daemon.sh 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390
  1. #!/bin/bash
  2. # filepath: ocr_platform/ocr_tools/daemons/paddleocr_local_daemon.sh
  3. # 对应: PaddleOCR-VL 本地 llama-server 服务(macOS),使用 GGUF 格式模型
  4. # 适用于 Mac M4 Pro 48G,使用 Metal GPU 加速
  5. # 模型下载地址: https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5-GGUF
  6. # unset https_proxy http_proxy HF_ENDPOINT
  7. # llama-server -hf PaddlePaddle/PaddleOCR-VL-1.5-GGUF
  8. # mv ~/Library/Caches/llama.cpp/PaddlePaddle_PaddleOCR-VL-1.5-GGUF_PaddleOCR-VL-1.5.gguf ~/models/paddleocr_vl
  9. # mv ~/Library/Caches/llama.cpp/PaddlePaddle_PaddleOCR-VL-1.5-GGUF_PaddleOCR-VL-1.5-mmproj.gguf ~/models/paddleocr_vl
  10. # curl -X POST http://localhost:8102/v1/chat/completions -d @payload.json
  11. LOGDIR="$HOME/workspace/logs"
  12. mkdir -p $LOGDIR
  13. PIDFILE="$LOGDIR/paddleocr_llamaserver.pid"
  14. LOGFILE="$LOGDIR/paddleocr_llamaserver.log"
  15. # 配置参数
  16. CONDA_ENV="mineru2"
  17. PORT="8102"
  18. HOST="0.0.0.0"
  19. # 本地 GGUF 模型路径(llama-server -hf 下载后的实际路径)
  20. HF_CACHE="$HOME/models/hf_home/hub/models--PaddlePaddle--PaddleOCR-VL-1.5-GGUF/snapshots/c8806b09a259ad6aaa7f401ed73dad2ff8df2c51"
  21. MODEL_PATH="$HF_CACHE/PaddleOCR-VL-1.5.gguf"
  22. MMPROJ_PATH="$HF_CACHE/PaddleOCR-VL-1.5-mmproj.gguf"
  23. # 模型别名(对外暴露的模型 ID,对应 yaml 中的 model_name)
  24. MODEL_NAME="PaddleOCR-VL-1.5"
  25. # llama-server 参数
  26. CONTEXT_SIZE="16384" # 上下文长度(需 >= max_tokens,推荐 8192-16384)
  27. GPU_LAYERS="99" # Metal GPU 层数(99 表示全部)
  28. THREADS="8" # CPU 线程数(M4 Pro 建议值)
  29. BATCH_SIZE="512" # 批处理大小
  30. UBATCH_SIZE="128" # 微批处理大小
  31. # conda 环境激活
  32. if [ -f "$HOME/anaconda3/etc/profile.d/conda.sh" ]; then
  33. source "$HOME/anaconda3/etc/profile.d/conda.sh"
  34. conda activate $CONDA_ENV
  35. elif [ -f "$HOME/miniconda3/etc/profile.d/conda.sh" ]; then
  36. source "$HOME/miniconda3/etc/profile.d/conda.sh"
  37. conda activate $CONDA_ENV
  38. elif [ -f "/opt/miniconda3/etc/profile.d/conda.sh" ]; then
  39. source /opt/miniconda3/etc/profile.d/conda.sh
  40. conda activate $CONDA_ENV
  41. else
  42. echo "Warning: conda initialization file not found, trying direct path"
  43. export PATH="/opt/miniconda3/envs/$CONDA_ENV/bin:$PATH"
  44. fi
  45. start() {
  46. if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
  47. echo "PaddleOCR-VL llama-server 已在运行"
  48. return 1
  49. fi
  50. echo "启动 PaddleOCR-VL llama-server 守护进程..."
  51. echo "Host: $HOST, Port: $PORT"
  52. echo "主模型: $MODEL_PATH"
  53. echo "多模态投影器: $MMPROJ_PATH"
  54. echo "上下文长度: $CONTEXT_SIZE"
  55. echo "GPU 层数: $GPU_LAYERS (Metal)"
  56. echo "线程数: $THREADS"
  57. # 检查模型文件是否存在
  58. if [ ! -f "$MODEL_PATH" ]; then
  59. echo "❌ 主模型文件不存在: $MODEL_PATH"
  60. echo "请确认模型已下载到 llama.cpp 缓存目录"
  61. return 1
  62. fi
  63. if [ ! -f "$MMPROJ_PATH" ]; then
  64. echo "❌ 多模态投影器文件不存在: $MMPROJ_PATH"
  65. echo "请确认 mmproj 文件已下载"
  66. return 1
  67. fi
  68. # 检查 llama-server 命令
  69. if ! command -v llama-server >/dev/null 2>&1; then
  70. echo "❌ llama-server 未找到"
  71. echo "请安装: brew install llama.cpp"
  72. return 1
  73. fi
  74. echo "🔧 使用 llama-server: $(which llama-server)"
  75. echo "🔧 llama.cpp 版本: $(llama-server --version 2>&1 | head -1 || echo 'Unknown')"
  76. echo "💻 系统信息:"
  77. echo " 架构: $(uname -m)"
  78. echo " 系统: $(uname -s)"
  79. echo " 内存: $(sysctl -n hw.memsize | awk '{printf "%.1f GB", $1/1024/1024/1024}')"
  80. # 启动 llama-server
  81. nohup llama-server \
  82. -m "$MODEL_PATH" \
  83. --mmproj "$MMPROJ_PATH" \
  84. --alias $MODEL_NAME \
  85. --host $HOST \
  86. --port $PORT \
  87. --media-path $HOME/workspace \
  88. -c $CONTEXT_SIZE \
  89. -ngl $GPU_LAYERS \
  90. -t $THREADS \
  91. -b $BATCH_SIZE \
  92. -ub $UBATCH_SIZE \
  93. --temp 0 \
  94. > $LOGFILE 2>&1 &
  95. echo $! > $PIDFILE
  96. echo "✅ PaddleOCR-VL llama-server 已启动,PID: $(cat $PIDFILE)"
  97. echo "📋 日志文件: $LOGFILE"
  98. echo "🌐 服务 URL: http://$HOST:$PORT"
  99. echo "📖 OpenAI 兼容 API: http://localhost:$PORT/v1 (chat/completions, models)"
  100. echo ""
  101. echo "等待服务启动..."
  102. sleep 5
  103. status
  104. }
  105. stop() {
  106. if [ ! -f $PIDFILE ]; then
  107. echo "PaddleOCR-VL llama-server 未在运行"
  108. return 1
  109. fi
  110. PID=$(cat $PIDFILE)
  111. echo "停止 PaddleOCR-VL llama-server (PID: $PID)..."
  112. kill $PID
  113. for i in {1..30}; do
  114. if ! kill -0 $PID 2>/dev/null; then
  115. break
  116. fi
  117. echo "等待进程停止... ($i/30)"
  118. sleep 1
  119. done
  120. if kill -0 $PID 2>/dev/null; then
  121. echo "强制终止进程..."
  122. kill -9 $PID
  123. fi
  124. rm -f $PIDFILE
  125. echo "✅ PaddleOCR-VL llama-server 已停止"
  126. }
  127. status() {
  128. if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
  129. PID=$(cat $PIDFILE)
  130. echo "✅ PaddleOCR-VL llama-server 正在运行 (PID: $PID)"
  131. echo "🌐 服务 URL: http://$HOST:$PORT"
  132. echo "📋 日志文件: $LOGFILE"
  133. # 检查端口监听状态
  134. if lsof -nP -iTCP:$PORT -sTCP:LISTEN >/dev/null 2>&1; then
  135. echo "🔗 端口 $PORT 正在监听"
  136. else
  137. echo "⚠️ 端口 $PORT 未在监听(服务可能正在启动)"
  138. fi
  139. # 检查 API 响应
  140. if command -v curl >/dev/null 2>&1; then
  141. if curl -s --connect-timeout 2 http://127.0.0.1:$PORT/v1/models > /dev/null 2>&1; then
  142. echo "🎯 API 响应正常"
  143. else
  144. echo "⚠️ API 无响应(服务可能正在启动)"
  145. fi
  146. fi
  147. # 显示进程内存使用
  148. if command -v ps >/dev/null 2>&1; then
  149. MEM=$(ps -o rss= -p $PID 2>/dev/null | awk '{printf "%.2f GB", $1/1024/1024}')
  150. if [ -n "$MEM" ]; then
  151. echo "💾 内存使用: $MEM"
  152. fi
  153. fi
  154. if [ -f $LOGFILE ]; then
  155. echo "📄 最近日志(最后 3 行):"
  156. tail -3 $LOGFILE | sed 's/^/ /'
  157. fi
  158. else
  159. echo "❌ PaddleOCR-VL llama-server 未在运行"
  160. if [ -f $PIDFILE ]; then
  161. echo "删除过期的 PID 文件..."
  162. rm -f $PIDFILE
  163. fi
  164. fi
  165. }
  166. logs() {
  167. if [ -f $LOGFILE ]; then
  168. echo "📄 PaddleOCR-VL llama-server 日志:"
  169. echo "====================="
  170. tail -f $LOGFILE
  171. else
  172. echo "❌ 日志文件不存在: $LOGFILE"
  173. fi
  174. }
  175. config() {
  176. echo "📋 当前配置:"
  177. echo " Conda 环境: $CONDA_ENV"
  178. echo " Host: $HOST"
  179. echo " Port: $PORT"
  180. echo " 模型别名: $MODEL_NAME"
  181. echo " 主模型路径: $MODEL_PATH"
  182. echo " 多模态投影器: $MMPROJ_PATH"
  183. echo " 上下文长度: $CONTEXT_SIZE"
  184. echo " GPU 层数: $GPU_LAYERS"
  185. echo " 线程数: $THREADS"
  186. echo " 批处理大小: $BATCH_SIZE"
  187. echo " 微批处理大小: $UBATCH_SIZE"
  188. echo " PID 文件: $PIDFILE"
  189. echo " 日志文件: $LOGFILE"
  190. echo ""
  191. echo "📦 模型文件检查:"
  192. if [ -f "$MODEL_PATH" ]; then
  193. SIZE=$(du -h "$MODEL_PATH" | cut -f1)
  194. echo " ✅ 主模型存在 ($SIZE)"
  195. else
  196. echo " ❌ 主模型不存在"
  197. fi
  198. if [ -f "$MMPROJ_PATH" ]; then
  199. SIZE=$(du -h "$MMPROJ_PATH" | cut -f1)
  200. echo " ✅ 多模态投影器存在 ($SIZE)"
  201. else
  202. echo " ❌ 多模态投影器不存在"
  203. fi
  204. echo ""
  205. echo "🔧 环境检查:"
  206. echo " llama-server: $(which llama-server 2>/dev/null || echo '未安装')"
  207. if command -v llama-server >/dev/null 2>&1; then
  208. LLAMA_VERSION=$(llama-server --version 2>&1 | head -1 || echo 'Unknown')
  209. echo " 版本: $LLAMA_VERSION"
  210. fi
  211. echo " Conda: $(which conda 2>/dev/null || echo '未找到')"
  212. echo " 当前 Python: $(which python 2>/dev/null || echo '未找到')"
  213. echo ""
  214. echo "💻 系统信息:"
  215. echo " 架构: $(uname -m)"
  216. echo " 系统版本: $(sw_vers -productVersion 2>/dev/null || echo 'Unknown')"
  217. echo " 总内存: $(sysctl -n hw.memsize 2>/dev/null | awk '{printf "%.1f GB", $1/1024/1024/1024}' || echo 'Unknown')"
  218. echo " CPU 核心: $(sysctl -n hw.ncpu 2>/dev/null || echo 'Unknown')"
  219. }
  220. test_api() {
  221. echo "🧪 测试 PaddleOCR-VL llama-server API..."
  222. if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
  223. echo "❌ PaddleOCR-VL llama-server 服务未在运行"
  224. return 1
  225. fi
  226. if ! command -v curl >/dev/null 2>&1; then
  227. echo "❌ curl 命令未找到"
  228. return 1
  229. fi
  230. echo "📡 测试 /v1/models 端点..."
  231. response=$(curl -s --connect-timeout 10 http://127.0.0.1:$PORT/v1/models)
  232. if [ $? -eq 0 ]; then
  233. echo "✅ Models 端点可访问"
  234. echo "$response" | python -m json.tool 2>/dev/null || echo "$response"
  235. else
  236. echo "❌ Models 端点不可访问"
  237. fi
  238. echo ""
  239. echo "📡 测试 /health 端点..."
  240. health=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/health)
  241. if [ $? -eq 0 ]; then
  242. echo "✅ Health 端点: $health"
  243. else
  244. echo "⚠️ Health 端点不可访问"
  245. fi
  246. }
  247. test_client() {
  248. echo "🧪 测试 PaddleOCR-VL 与 llama-server 集成..."
  249. if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
  250. echo "❌ PaddleOCR-VL llama-server 服务未在运行,请先启动: $0 start"
  251. return 1
  252. fi
  253. CONFIG_FILE="/Users/zhch158/workspace/repository.git/ocr_platform/ocr_tools/universal_doc_parser/config/bank_statement_paddleocr_local.yaml"
  254. echo "📄 配置文件: $CONFIG_FILE"
  255. echo ""
  256. echo "确保配置文件中 vl_recognition.api_url 指向: http://localhost:$PORT/v1/chat/completions"
  257. echo ""
  258. echo "测试命令示例:"
  259. echo " cd /Users/zhch158/workspace/repository.git/ocr_platform/ocr_tools/universal_doc_parser"
  260. echo " conda activate mineru2"
  261. echo " python parse.py --input /path/to/test/image.png --config $CONFIG_FILE --debug"
  262. echo ""
  263. echo "或者使用 curl 直接测试 API:"
  264. echo " curl -X POST http://localhost:$PORT/v1/chat/completions \\"
  265. echo " -H 'Content-Type: application/json' \\"
  266. echo " -d '{"
  267. echo " \"model\": \"paddleocr-vl\","
  268. echo " \"messages\": ["
  269. echo " {"
  270. echo " \"role\": \"user\","
  271. echo " \"content\": ["
  272. echo " {\"type\": \"text\", \"text\": \"Table Recognition:\"},"
  273. echo " {\"type\": \"image_url\", \"image_url\": {\"url\": \"file:///path/to/image.png\"}}"
  274. echo " ]"
  275. echo " }"
  276. echo " ],"
  277. echo " \"max_tokens\": 4096"
  278. echo " }'"
  279. }
  280. usage() {
  281. echo "PaddleOCR-VL llama-server 服务守护进程(macOS)"
  282. echo "==========================================="
  283. echo "用法: $0 {start|stop|restart|status|logs|config|test|test-client}"
  284. echo ""
  285. echo "命令:"
  286. echo " start - 启动 PaddleOCR-VL llama-server 服务"
  287. echo " stop - 停止 PaddleOCR-VL llama-server 服务"
  288. echo " restart - 重启 PaddleOCR-VL llama-server 服务"
  289. echo " status - 显示服务状态和资源使用"
  290. echo " logs - 显示服务日志(跟踪模式)"
  291. echo " config - 显示当前配置"
  292. echo " test - 测试 /v1/models API 端点"
  293. echo " test-client - 显示如何测试与配置文件集成"
  294. echo ""
  295. echo "配置(编辑脚本修改):"
  296. echo " Host: $HOST"
  297. echo " Port: $PORT"
  298. echo " 主模型: $MODEL_PATH"
  299. echo " 多模态投影器: $MMPROJ_PATH"
  300. echo " 上下文长度: $CONTEXT_SIZE"
  301. echo " GPU 层数: $GPU_LAYERS (Metal)"
  302. echo ""
  303. echo "示例:"
  304. echo " ./paddleocr_local_daemon.sh start"
  305. echo " ./paddleocr_local_daemon.sh status"
  306. echo " ./paddleocr_local_daemon.sh logs"
  307. echo " ./paddleocr_local_daemon.sh test"
  308. echo ""
  309. echo "前置要求:"
  310. echo " 1. 安装 llama.cpp: brew install llama.cpp"
  311. echo " 2. 模型文件位于: ~/Library/Caches/llama.cpp/"
  312. echo " 3. conda 环境 mineru2 已配置"
  313. }
  314. case "$1" in
  315. start)
  316. start
  317. ;;
  318. stop)
  319. stop
  320. ;;
  321. restart)
  322. stop
  323. sleep 3
  324. start
  325. ;;
  326. status)
  327. status
  328. ;;
  329. logs)
  330. logs
  331. ;;
  332. config)
  333. config
  334. ;;
  335. test)
  336. test_api
  337. ;;
  338. test-client)
  339. test_client
  340. ;;
  341. *)
  342. usage
  343. exit 1
  344. ;;
  345. esac