chore(deploy): archive v3.2 and develop deployment artifacts
parent
4b4079d96f
commit
e34d9011ed
|
|
@ -0,0 +1,6 @@
|
|||
*.tar
|
||||
*.tar.gz
|
||||
.env
|
||||
*.env
|
||||
!.env.example
|
||||
releases/
|
||||
|
|
@ -0,0 +1,8 @@
|
|||
# 旧部署材料归档
|
||||
|
||||
本目录从工作区原 deploy_artifacts 原样归档,对应 MaxKB v3.2 / markitdown develop。
|
||||
SHA256SUMS.txt 属于当时离线发布包,列出的镜像和模型包未纳入 Git。
|
||||
|
||||
这是旧版部署记录。start-all.sh 含递归修改权限,文档含全局清理示例;不要用于日常增量升级。
|
||||
升级 v3.2-refact / develop-refact 请使用重构分支 deploy_artifacts 中的升级流程。
|
||||
新流程复用原持久化目录,备份后执行数据库迁移,保留旧镜像和失败现场。
|
||||
|
|
@ -0,0 +1,481 @@
|
|||
# 知识库问答助手与思维导图助手 Docker 部署说明
|
||||
|
||||
本文档用于在测试环境或生产环境部署已打包好的知识库问答助手、思维导图助手及统一网关。
|
||||
|
||||
## 1. 下载安装包
|
||||
|
||||
将开发机 `/home/tlw/unis/deploy_artifacts` 下的部署产物下载或拷贝到目标服务器,例如:
|
||||
|
||||
```bash
|
||||
mkdir -p /data1/zywlw/tar
|
||||
cd /data1/zywlw/tar
|
||||
```
|
||||
|
||||
目标目录中建议包含以下文件:
|
||||
|
||||
```text
|
||||
docker-compose.unis.yml
|
||||
start-all.sh
|
||||
load-images.sh
|
||||
download-models.sh
|
||||
SHA256SUMS.txt
|
||||
README_DEPLOY.md
|
||||
DEPLOY_CN.md
|
||||
unis-kb-latest.tar
|
||||
unis-gateway-latest.tar
|
||||
unis-mindmap-api-latest.tar
|
||||
unis-mindmap-api-corex-latest.tar
|
||||
unis-mindmap-api-cuda-latest.tar
|
||||
unis-mindmap-web-latest.tar
|
||||
mineru-models-modelscope.tar.gz
|
||||
```
|
||||
|
||||
其中:
|
||||
|
||||
- `unis-kb-latest.tar`:知识库问答助手镜像。
|
||||
- `unis-gateway-latest.tar`:统一入口网关镜像,对外提供 `8989` 端口。
|
||||
- `unis-mindmap-api-latest.tar`:思维导图助手 MinerU/API 默认镜像,通常对应天数/CoreX 版本。
|
||||
- `unis-mindmap-api-corex-latest.tar`:思维导图助手 MinerU/API 天数/CoreX GPU 版本。
|
||||
- `unis-mindmap-api-cuda-latest.tar`:思维导图助手 MinerU/API NVIDIA CUDA GPU 版本。
|
||||
- `unis-mindmap-web-latest.tar`:思维导图助手 Web 镜像,对外提供 `9898` 端口。
|
||||
- `mineru-models-modelscope.tar.gz`:思维导图助手模型包,可独立更新。
|
||||
|
||||
## 2. 检查资源完整性
|
||||
|
||||
进入部署目录后,先校验文件完整性:
|
||||
|
||||
```bash
|
||||
cd /data1/zywlw/tar
|
||||
sha256sum -c SHA256SUMS.txt
|
||||
```
|
||||
|
||||
所有文件显示 `成功` 或 `OK` 后再继续部署。
|
||||
|
||||
如果提示某个文件校验失败,需要重新拷贝对应文件,不建议继续使用。
|
||||
|
||||
部署前还建议检查 Docker 存储空间:
|
||||
|
||||
```bash
|
||||
df -h
|
||||
docker system df
|
||||
```
|
||||
|
||||
`unis-mindmap-api-*-latest.tar` 和 `unis-kb-latest.tar` 较大,镜像加载时还会产生临时写入。建议 Docker 数据盘至少预留 `15G - 20G` 可用空间。如果同时加载天数/CoreX 和 NVIDIA/CUDA 两个 API 镜像,需要预留更多空间。
|
||||
|
||||
如果加载镜像时出现 `no space left on device`,需要先清理旧镜像、旧容器或扩容 Docker 数据目录:
|
||||
|
||||
```bash
|
||||
docker container prune -f
|
||||
docker image prune -af
|
||||
docker builder prune -af
|
||||
docker system prune -af
|
||||
```
|
||||
|
||||
清理前请确认没有仍需保留的业务容器或镜像。
|
||||
|
||||
## 3. 解压模型文件
|
||||
|
||||
思维导图助手的模型文件不打进 API 镜像,需要单独放到宿主机目录,并通过 `MINERU_MODEL_DIR` 挂载进容器。
|
||||
|
||||
推荐模型目录:
|
||||
|
||||
```bash
|
||||
/data1/zywlw/mineru-models
|
||||
```
|
||||
|
||||
首次部署或模型包有更新时执行:
|
||||
|
||||
```bash
|
||||
mkdir -p /data1/zywlw/mineru-models
|
||||
tar -xzf mineru-models-modelscope.tar.gz -C /data1/zywlw
|
||||
```
|
||||
|
||||
解压后目录结构通常为:
|
||||
|
||||
```text
|
||||
/data1/zywlw/mineru-models/pipeline
|
||||
/data1/zywlw/mineru-models/vlm
|
||||
```
|
||||
|
||||
如果目标目录已有旧模型,可以覆盖更新:
|
||||
|
||||
```bash
|
||||
tar -xzf mineru-models-modelscope.tar.gz -C /data1/zywlw --overwrite
|
||||
```
|
||||
|
||||
模型包只在模型有变化时需要更新。普通代码镜像升级时,可以复用已有模型目录。
|
||||
|
||||
可检查关键模型文件是否存在:
|
||||
|
||||
```bash
|
||||
find /data1/zywlw/mineru-models -name 'yolo_v8_ft.pt'
|
||||
find /data1/zywlw/mineru-models -name 'doclayout_yolo_docstructbench_imgsz1280_2501.pt'
|
||||
find /data1/zywlw/mineru-models -name 'model.safetensors'
|
||||
```
|
||||
|
||||
## 4. 加载镜像
|
||||
|
||||
按目标服务器 GPU 类型选择加载方式。
|
||||
|
||||
天数/CoreX GPU 服务器执行:
|
||||
|
||||
```bash
|
||||
cd /data1/zywlw/tar
|
||||
chmod +x *.sh
|
||||
GPU_FLAVOR=corex ./load-images.sh
|
||||
```
|
||||
|
||||
NVIDIA CUDA GPU 服务器执行:
|
||||
|
||||
```bash
|
||||
cd /data1/zywlw/tar
|
||||
chmod +x *.sh
|
||||
GPU_FLAVOR=cuda ./load-images.sh
|
||||
```
|
||||
|
||||
`GPU_FLAVOR` 说明:
|
||||
|
||||
- `corex`:加载天数/CoreX 版本 `unis-mindmap-api-corex-latest.tar`。
|
||||
- `cuda`:加载 NVIDIA CUDA 版本 `unis-mindmap-api-cuda-latest.tar`。
|
||||
|
||||
脚本会将选中的 API 镜像标记为 `unis-mindmap-api:latest`,因此后续 `./start-all.sh` 不需要再区分 GPU 类型。
|
||||
|
||||
脚本会依次加载:
|
||||
|
||||
```text
|
||||
unis-kb:latest
|
||||
unis-gateway:latest
|
||||
unis-mindmap-web:latest
|
||||
unis-mindmap-api:latest
|
||||
```
|
||||
|
||||
加载完成后可以检查镜像:
|
||||
|
||||
```bash
|
||||
docker images | grep -E 'unis-kb|unis-gateway|unis-mindmap'
|
||||
```
|
||||
|
||||
如果服务器需要 `sudo` 才能执行 Docker 命令,可以使用:
|
||||
|
||||
```bash
|
||||
sudo GPU_FLAVOR=corex ./load-images.sh
|
||||
# 或
|
||||
sudo GPU_FLAVOR=cuda ./load-images.sh
|
||||
```
|
||||
|
||||
## 5. 启动服务
|
||||
|
||||
推荐启动命令如下:
|
||||
|
||||
```bash
|
||||
cd /data1/zywlw/tar
|
||||
|
||||
MINDMAP_LLM_BASE_URL=http://10.100.53.199:9527/v1 \
|
||||
MINDMAP_LLM_API_KEY=xxx \
|
||||
MINDMAP_LLM_MODEL=gemma-4-26B \
|
||||
MINERU_MODEL_DIR=/data1/zywlw/mineru-models \
|
||||
KB_DATA_DIR=/data1/zywlw/maxkb \
|
||||
MINDMAP_PUBLIC_URL=/mindmap/ \
|
||||
MINERU_DOCKER_RUNTIME=nvidia \
|
||||
MINERU_CUDA_VISIBLE_DEVICES=1 \
|
||||
MINERU_GPU_MEMORY_UTILIZATION=0.75 \
|
||||
./start-all.sh
|
||||
```
|
||||
|
||||
启动后会创建并运行以下容器:
|
||||
|
||||
```text
|
||||
airag 知识库问答助手
|
||||
mineru-api 思维导图解析 API
|
||||
mindmap-web 思维导图 Web 页面
|
||||
unis-gateway 统一访问网关
|
||||
```
|
||||
|
||||
访问地址:
|
||||
|
||||
```text
|
||||
知识库后台:http://服务器IP:8989/admin/
|
||||
知识库聊天:http://服务器IP:8989/chat/
|
||||
思维导图,经网关访问:http://服务器IP:8989/mindmap/
|
||||
思维导图,直接端口访问:http://服务器IP:9898/
|
||||
```
|
||||
|
||||
## 6. 重要启动参数说明
|
||||
|
||||
### MINDMAP_LLM_BASE_URL
|
||||
|
||||
智能整理、智能总结使用的大模型服务地址。
|
||||
|
||||
需要填写 OpenAI 兼容接口的 base URL,不要带 `/chat/completions`。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
MINDMAP_LLM_BASE_URL=http://10.100.53.199:9527/v1
|
||||
```
|
||||
|
||||
如果实际接口是:
|
||||
|
||||
```text
|
||||
http://10.100.53.199:9527/v1/chat/completions
|
||||
```
|
||||
|
||||
则参数应配置为:
|
||||
|
||||
```text
|
||||
http://10.100.53.199:9527/v1
|
||||
```
|
||||
|
||||
如果不配置该参数,思维导图助手的文件转换仍可使用,但智能整理/智能总结会提示未配置模型服务。
|
||||
|
||||
### MINDMAP_LLM_API_KEY
|
||||
|
||||
大模型服务的 API Key。
|
||||
|
||||
如果本地模型服务不校验 Key,可以填任意非空值,例如:
|
||||
|
||||
```bash
|
||||
MINDMAP_LLM_API_KEY=EMPTY
|
||||
```
|
||||
|
||||
如果服务有鉴权要求,需要填写真实 Key。
|
||||
|
||||
### MINDMAP_LLM_MODEL
|
||||
|
||||
智能整理、智能总结调用的模型名称,需要与大模型服务端提供的模型名一致。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
MINDMAP_LLM_MODEL=gemma-4-26B
|
||||
```
|
||||
|
||||
### MINERU_MODEL_DIR
|
||||
|
||||
宿主机上的思维导图模型目录,会挂载到 `mineru-api` 容器内的 `/models`。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
MINERU_MODEL_DIR=/data1/zywlw/mineru-models
|
||||
```
|
||||
|
||||
该目录中应包含:
|
||||
|
||||
```text
|
||||
pipeline/
|
||||
vlm/
|
||||
```
|
||||
|
||||
建议放在数据盘,避免系统盘空间不足。
|
||||
|
||||
### KB_DATA_DIR
|
||||
|
||||
知识库问答助手的数据持久化目录,会挂载到容器内 `/opt/maxkb`。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
KB_DATA_DIR=/data1/zywlw/maxkb
|
||||
```
|
||||
|
||||
该目录会保存知识库运行数据、数据库数据、上传文件等持久化内容。升级镜像时不要删除该目录。
|
||||
|
||||
### MINDMAP_PUBLIC_URL
|
||||
|
||||
知识库问答助手中打开思维导图助手时使用的访问地址。
|
||||
|
||||
通过统一网关访问时使用:
|
||||
|
||||
```bash
|
||||
MINDMAP_PUBLIC_URL=/mindmap/
|
||||
```
|
||||
|
||||
此时访问链路为:
|
||||
|
||||
```text
|
||||
http://服务器IP:8989/mindmap/
|
||||
```
|
||||
|
||||
如果希望知识库菜单直接打开思维导图独立端口,也可以配置为完整地址:
|
||||
|
||||
```bash
|
||||
MINDMAP_PUBLIC_URL=http://服务器IP:9898/
|
||||
```
|
||||
|
||||
### MINERU_CUDA_VISIBLE_DEVICES
|
||||
|
||||
指定 `mineru-api` 容器使用哪张 GPU。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
MINERU_CUDA_VISIBLE_DEVICES=1
|
||||
```
|
||||
|
||||
表示使用宿主机上的 GPU 1。
|
||||
|
||||
如果 GPU 0 已被其它大模型服务占用,建议配置为 `1` 或其它空闲 GPU。
|
||||
|
||||
天数 GPU 服务器可使用:
|
||||
|
||||
```bash
|
||||
ixsmi
|
||||
```
|
||||
|
||||
查看显存占用。
|
||||
|
||||
NVIDIA GPU 服务器可使用:
|
||||
|
||||
```bash
|
||||
nvidia-smi
|
||||
```
|
||||
|
||||
查看显存占用。
|
||||
|
||||
### MINERU_DOCKER_RUNTIME
|
||||
|
||||
指定 `mineru-api` 容器使用的 Docker runtime。
|
||||
|
||||
天数/CoreX 环境通常不需要设置该参数,默认使用 `runc`。
|
||||
|
||||
NVIDIA 环境如果 `docker info` 中可以看到 `nvidia` runtime,需要设置:
|
||||
|
||||
```bash
|
||||
MINERU_DOCKER_RUNTIME=nvidia
|
||||
```
|
||||
|
||||
如果不设置,宿主机 `nvidia-smi` 即使正常,容器内也可能看不到 `libcuda.so.1`,从而报:
|
||||
|
||||
```text
|
||||
CUDA not available
|
||||
Device string must not be empty
|
||||
```
|
||||
|
||||
### MINERU_GPU_MEMORY_UTILIZATION
|
||||
|
||||
控制 VLLM/MinerU 使用 GPU 显存的比例,默认值为 `0.5`。
|
||||
|
||||
示例:
|
||||
|
||||
```bash
|
||||
MINERU_GPU_MEMORY_UTILIZATION=0.4
|
||||
```
|
||||
|
||||
当同一张 GPU 上还有其它服务,或启动时出现显存不足,可以调低该值。
|
||||
|
||||
如果是 8G 左右的 NVIDIA 显卡,建议从 `0.75` 开始尝试。该值太低时,VLLM 可能没有足够 KV cache 空间,并报:
|
||||
|
||||
```text
|
||||
No available memory for the cache blocks
|
||||
```
|
||||
|
||||
### MINDMAP_LLM_TIMEOUT
|
||||
|
||||
智能整理、智能总结请求大模型服务的超时时间,单位秒,默认 `180`。
|
||||
|
||||
文档较大或模型响应较慢时可以调大:
|
||||
|
||||
```bash
|
||||
MINDMAP_LLM_TIMEOUT=300
|
||||
```
|
||||
|
||||
## 7. 常用运维命令
|
||||
|
||||
查看容器状态:
|
||||
|
||||
```bash
|
||||
docker ps
|
||||
```
|
||||
|
||||
查看知识库助手日志:
|
||||
|
||||
```bash
|
||||
docker logs -f airag
|
||||
```
|
||||
|
||||
查看思维导图 API 日志:
|
||||
|
||||
```bash
|
||||
docker logs -f mineru-api
|
||||
```
|
||||
|
||||
查看网关日志:
|
||||
|
||||
```bash
|
||||
docker logs -f unis-gateway
|
||||
```
|
||||
|
||||
停止服务:
|
||||
|
||||
```bash
|
||||
docker compose -f docker-compose.unis.yml down
|
||||
```
|
||||
|
||||
重新启动:
|
||||
|
||||
```bash
|
||||
./start-all.sh
|
||||
```
|
||||
|
||||
升级镜像后强制重建容器:
|
||||
|
||||
```bash
|
||||
docker compose -f docker-compose.unis.yml up -d --force-recreate
|
||||
```
|
||||
|
||||
## 8. 常见问题
|
||||
|
||||
### 首页加载报 `ERR_INCOMPLETE_CHUNKED_ENCODING`
|
||||
|
||||
一般优先检查是否加载了最新镜像,或者浏览器缓存了旧的前端入口文件。
|
||||
|
||||
检查首页实际引用的 JS:
|
||||
|
||||
```bash
|
||||
curl -s http://服务器IP:8989/admin/ | grep -o 'assets/admin-[^"]*\.js'
|
||||
docker exec airag grep -o 'assets/admin-[^"]*\.js' /opt/maxkb-app/apps/static/admin/index.html
|
||||
```
|
||||
|
||||
如果浏览器仍请求旧文件,尝试无痕窗口或清理浏览器缓存。
|
||||
|
||||
### 加载镜像时报 `no space left on device`
|
||||
|
||||
说明 Docker 数据目录空间不足。先检查:
|
||||
|
||||
```bash
|
||||
df -h
|
||||
docker system df
|
||||
```
|
||||
|
||||
再按需清理旧镜像和构建缓存:
|
||||
|
||||
```bash
|
||||
docker container prune -f
|
||||
docker image prune -af
|
||||
docker builder prune -af
|
||||
docker system prune -af
|
||||
```
|
||||
|
||||
### 思维导图提示缺少模型文件
|
||||
|
||||
检查模型目录挂载是否正确:
|
||||
|
||||
```bash
|
||||
docker exec mineru-api ls -lah /models
|
||||
docker exec mineru-api find /models -name 'yolo_v8_ft.pt'
|
||||
```
|
||||
|
||||
如果找不到关键模型文件,需要重新解压或更新 `mineru-models-modelscope.tar.gz`。
|
||||
|
||||
### 智能整理提示未配置模型服务
|
||||
|
||||
检查启动时是否设置:
|
||||
|
||||
```bash
|
||||
MINDMAP_LLM_BASE_URL
|
||||
MINDMAP_LLM_API_KEY
|
||||
MINDMAP_LLM_MODEL
|
||||
```
|
||||
|
||||
其中 `MINDMAP_LLM_BASE_URL` 应为 `/v1` 级别地址,不要带 `/chat/completions`。
|
||||
|
|
@ -0,0 +1,58 @@
|
|||
# Unis AI-RAG / MindMap Docker deployment
|
||||
|
||||
## Images
|
||||
|
||||
- `unis-kb-latest.tar`: knowledge-base assistant, image `unis-kb:latest`
|
||||
- `unis-gateway-latest.tar`: unified gateway, image `unis-gateway:latest`
|
||||
- `unis-mindmap-api-latest.tar`: MindMap API runtime image, image `unis-mindmap-api:latest`
|
||||
- `unis-mindmap-web-latest.tar`: MindMap Web UI, image `unis-mindmap-web:latest`
|
||||
|
||||
The MindMap API image does not contain MinerU model weights. Mount or download models separately.
|
||||
|
||||
## Routes
|
||||
|
||||
- Knowledge assistant: `http://SERVER_IP:8989/admin/`
|
||||
- Chat page: `http://SERVER_IP:8989/chat/`
|
||||
- MindMap via knowledge assistant port: `http://SERVER_IP:8989/mindmap/`
|
||||
- MindMap direct access: `http://SERVER_IP:9898/`
|
||||
|
||||
## Deploy
|
||||
|
||||
```bash
|
||||
sha256sum -c SHA256SUMS.txt
|
||||
./load-images.sh
|
||||
SOURCE=modelscope MINERU_MODEL_DIR=/data/unis/mineru-models ./download-models.sh
|
||||
MINDMAP_PUBLIC_URL=/mindmap/ KB_DATA_DIR=/opt/maxkb MINERU_MODEL_DIR=/data/unis/mineru-models ./start-all.sh
|
||||
```
|
||||
|
||||
`MINDMAP_PUBLIC_URL=/mindmap/` makes the knowledge assistant open MindMap through the gateway path. You can also set it to `http://SERVER_IP:9898/` if you want the menu to open the direct MindMap port.
|
||||
|
||||
If Docker requires root permission on the target server, run the commands with `sudo` or keep `sudo` in `start-all.sh`.
|
||||
|
||||
For an NVIDIA CUDA build, build the API image on a CUDA-capable build machine:
|
||||
|
||||
```bash
|
||||
cd UnisMindMap
|
||||
API_FLAVOR=cuda VERSION=$(date +%Y%m%d%H%M%S) ./scripts/build-images.sh
|
||||
```
|
||||
|
||||
For a Tianshu/CoreX build:
|
||||
|
||||
```bash
|
||||
cd UnisMindMap
|
||||
API_FLAVOR=corex VERSION=$(date +%Y%m%d%H%M%S) ./scripts/build-images.sh
|
||||
```
|
||||
|
||||
## Upload
|
||||
|
||||
```bash
|
||||
TARGET=root@TEST_SERVER TARGET_DIR=/opt/unis/deploy_artifacts ./upload-artifacts.sh
|
||||
```
|
||||
|
||||
## Notes
|
||||
|
||||
- `KB_DATA_DIR` is the persistent data directory for the knowledge-base assistant. Put it on a data disk in production.
|
||||
- `MINERU_MODEL_DIR` is the persistent model directory for the MindMap assistant. Put it on a data disk and reuse it across image upgrades.
|
||||
- The gateway container is the only public reverse proxy needed for `/admin/`, `/chat/`, and `/mindmap/`. MindMap Web also exposes port `9898` for direct access.
|
||||
- Use `MINERU_GPU_MEMORY_UTILIZATION=0.3` or another value below `1` when a GPU has limited free memory.
|
||||
- The MindMap API has separate Dockerfiles for CoreX/Tianshu and NVIDIA CUDA. Keep the same container name `mineru-api` and port `8000`.
|
||||
|
|
@ -0,0 +1,15 @@
|
|||
300c2703333f8303ea1b251c2c844100f18b4c7ddffdd5c22d578e80649e794e unis-gateway-latest.tar
|
||||
73392cea6dadc05a6024187fe18b90cd2e1c85ad696c270e9fee91650d6958bf unis-kb-latest.tar
|
||||
8c41177ab73119d6032254fe4c2d7af3eae81436df4e34cbfd2ac032bbdfc1b7 unis-mindmap-api-corex-latest.tar
|
||||
a42c38781c5a6aa3ff104bc95876cd60888d45b84bd155cfa72b688dc24948a7 unis-mindmap-api-cuda-latest.tar
|
||||
384bcdfbb4eb6885e21db5307d0c9540a3637b36d3d5aaeaf965a53b9ddf4629 unis-mindmap-api-latest.tar
|
||||
a0089573e7ae940dae82a37a5229f9ddc0b88674e244fd2ebbc157cbf0b66860 unis-mindmap-web-latest.tar
|
||||
0f029d8d4de80086853c5e9c648ba4a377e2064ad5a7eb0a5aa635ea85bed72d mineru-models-modelscope.tar.gz
|
||||
ea0ccf6e41e1af8dcff994c2d244d22e97dc79963e0ebad10051748a36c191b8 download-models.sh
|
||||
0bbcfd3325c0f03152541ca3b7212acc2afe109844dd2f331578ea27be214ccc load-images.sh
|
||||
4dad5acf494504af9220629bd357fb51b821eb60cc723a511fc227d4adbca0b9 save-images.sh
|
||||
b9f65892123f2a1eec447488bdb660ba3d88fd6e37b607b9d1467a3170c7cd33 start-all.sh
|
||||
5934edd4ecd97c403582a958670b7a1d33248bc8d4e04df9da7f3337ecd3df9b upload-artifacts.sh
|
||||
875f21227eb3e3ef6099c5157ac44b2ee0123afcc8e633052a08cb030094b4ae README_DEPLOY.md
|
||||
5d0763a3af810aeb59aac1d23b508dfd12dc772b624dcc9874a291746949e8b3 DEPLOY_CN.md
|
||||
c1a1d833e18ec8780aaba958f9da775ac43ac365da80283429e9ad00a6d53dda docker-compose.unis.yml
|
||||
|
|
@ -0,0 +1,76 @@
|
|||
services:
|
||||
mineru-api:
|
||||
image: ${MINDMAP_API_IMAGE:-unis-mindmap-api:latest}
|
||||
container_name: mineru-api
|
||||
restart: unless-stopped
|
||||
runtime: ${MINERU_DOCKER_RUNTIME:-runc}
|
||||
privileged: true
|
||||
ipc: host
|
||||
shm_size: 32g
|
||||
volumes:
|
||||
- ${MINERU_MODEL_DIR:-/data/unis/mineru-models}:/models
|
||||
environment:
|
||||
MINERU_MODEL_SOURCE: ${MINERU_MODEL_SOURCE:-local}
|
||||
MINERU_MODEL_DIR: /models
|
||||
MINERU_PIPELINE_MODEL_DIR: ${MINERU_CONTAINER_PIPELINE_MODEL_DIR:-/models/pipeline}
|
||||
MINERU_VLM_MODEL_DIR: ${MINERU_CONTAINER_VLM_MODEL_DIR:-/models/vlm}
|
||||
PYTHONPATH: ${MINERU_PYTHONPATH:-/app:/usr/local/corex-4.4.0/lib64/python3/dist-packages}
|
||||
MINERU_GPU_MEMORY_UTILIZATION: ${MINERU_GPU_MEMORY_UTILIZATION:-0.5}
|
||||
NVIDIA_VISIBLE_DEVICES: ${MINERU_NVIDIA_VISIBLE_DEVICES:-all}
|
||||
NVIDIA_DRIVER_CAPABILITIES: ${MINERU_NVIDIA_DRIVER_CAPABILITIES:-compute,utility}
|
||||
CUDA_VISIBLE_DEVICES: ${MINERU_CUDA_VISIBLE_DEVICES:-1}
|
||||
MINDMAP_LLM_BASE_URL: ${MINDMAP_LLM_BASE_URL:-}
|
||||
MINDMAP_LLM_API_KEY: ${MINDMAP_LLM_API_KEY:-}
|
||||
MINDMAP_LLM_MODEL: ${MINDMAP_LLM_MODEL:-gemma-4-26B}
|
||||
MINDMAP_LLM_TIMEOUT: ${MINDMAP_LLM_TIMEOUT:-180}
|
||||
MINDMAP_LLM_TEMPERATURE: ${MINDMAP_LLM_TEMPERATURE:-0.2}
|
||||
MINDMAP_LLM_MAX_CONTEXT_TOKENS: ${MINDMAP_LLM_MAX_CONTEXT_TOKENS:-32768}
|
||||
MINDMAP_LLM_MAX_OUTPUT_TOKENS: ${MINDMAP_LLM_MAX_OUTPUT_TOKENS:-4096}
|
||||
MINDMAP_LLM_SAFETY_TOKENS: ${MINDMAP_LLM_SAFETY_TOKENS:-1024}
|
||||
command: >
|
||||
mineru-api --host 0.0.0.0 --port 8000
|
||||
networks:
|
||||
- unis-ai
|
||||
|
||||
mindmap-web:
|
||||
image: unis-mindmap-web:latest
|
||||
container_name: mindmap-web
|
||||
restart: unless-stopped
|
||||
ports:
|
||||
- "9898:9898"
|
||||
environment:
|
||||
MINERU_API_URL: http://mineru-api:8000
|
||||
depends_on:
|
||||
- mineru-api
|
||||
networks:
|
||||
- unis-ai
|
||||
|
||||
airag:
|
||||
image: unis-kb:latest
|
||||
container_name: airag
|
||||
restart: unless-stopped
|
||||
volumes:
|
||||
- ${KB_DATA_DIR:-/opt/maxkb}:/opt/maxkb
|
||||
environment:
|
||||
MAXKB_MINDMAP_URL: ${MINDMAP_PUBLIC_URL:-/mindmap/}
|
||||
networks:
|
||||
- unis-ai
|
||||
|
||||
gateway:
|
||||
image: unis-gateway:latest
|
||||
container_name: unis-gateway
|
||||
restart: unless-stopped
|
||||
ports:
|
||||
- "8989:8989"
|
||||
environment:
|
||||
KB_UPSTREAM: http://airag:8080
|
||||
MINDMAP_UPSTREAM: http://mindmap-web:9898
|
||||
depends_on:
|
||||
- airag
|
||||
- mindmap-web
|
||||
networks:
|
||||
- unis-ai
|
||||
|
||||
networks:
|
||||
unis-ai:
|
||||
name: unis-ai
|
||||
|
|
@ -0,0 +1,76 @@
|
|||
#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
IMAGE=${MINDMAP_API_IMAGE:-unis-mindmap-api:latest}
|
||||
SOURCE=${SOURCE:-modelscope}
|
||||
MODEL_TYPE=${MODEL_TYPE:-all}
|
||||
MINERU_MODEL_DIR=${MINERU_MODEL_DIR:-/data/unis/mineru-models}
|
||||
|
||||
sudo mkdir -p "$MINERU_MODEL_DIR"/pipeline "$MINERU_MODEL_DIR"/vlm
|
||||
sudo chmod -R 777 "$MINERU_MODEL_DIR"
|
||||
|
||||
sudo docker run --rm -i \
|
||||
-e SOURCE="$SOURCE" \
|
||||
-e MODEL_TYPE="$MODEL_TYPE" \
|
||||
-v "${MINERU_MODEL_DIR}:/models" \
|
||||
"$IMAGE" \
|
||||
python3 - <<'PY'
|
||||
import os
|
||||
from mineru.utils.enum_class import ModelPath
|
||||
|
||||
source = os.environ.get("SOURCE", "modelscope")
|
||||
model_type = os.environ.get("MODEL_TYPE", "all")
|
||||
|
||||
if source == "modelscope":
|
||||
from modelscope import snapshot_download
|
||||
pipeline_repo = ModelPath.pipeline_root_modelscope
|
||||
vlm_repo = ModelPath.vlm_root_modelscope
|
||||
elif source == "huggingface":
|
||||
from huggingface_hub import snapshot_download
|
||||
pipeline_repo = ModelPath.pipeline_root_hf
|
||||
vlm_repo = ModelPath.vlm_root_hf
|
||||
else:
|
||||
raise SystemExit(f"Unsupported SOURCE={source}; use modelscope or huggingface")
|
||||
|
||||
pipeline_dir = "/models/pipeline"
|
||||
vlm_dir = "/models/vlm"
|
||||
pipeline_paths = [
|
||||
ModelPath.doclayout_yolo,
|
||||
ModelPath.yolo_v8_mfd,
|
||||
ModelPath.unimernet_small,
|
||||
ModelPath.pytorch_paddle,
|
||||
ModelPath.layout_reader,
|
||||
ModelPath.slanet_plus,
|
||||
ModelPath.unet_structure,
|
||||
ModelPath.paddle_table_cls,
|
||||
ModelPath.paddle_orientation_classification,
|
||||
ModelPath.pp_formulanet_plus_m,
|
||||
]
|
||||
|
||||
def download(repo, local_dir, allow_patterns=None):
|
||||
kwargs = {"local_dir": local_dir}
|
||||
if allow_patterns:
|
||||
kwargs["allow_patterns"] = allow_patterns
|
||||
try:
|
||||
return snapshot_download(repo, **kwargs)
|
||||
except TypeError:
|
||||
kwargs.pop("local_dir")
|
||||
return snapshot_download(repo, **kwargs)
|
||||
|
||||
if model_type in ("pipeline", "all"):
|
||||
patterns = []
|
||||
for path in pipeline_paths:
|
||||
rel = path.strip("/")
|
||||
patterns.extend([rel, f"{rel}/*"])
|
||||
print(f"Downloading pipeline models from {source} to {pipeline_dir}")
|
||||
download(pipeline_repo, pipeline_dir, patterns)
|
||||
|
||||
if model_type in ("vlm", "all"):
|
||||
print(f"Downloading VLM models from {source} to {vlm_dir}")
|
||||
download(vlm_repo, vlm_dir)
|
||||
|
||||
if model_type not in ("pipeline", "vlm", "all"):
|
||||
raise SystemExit(f"Unsupported MODEL_TYPE={model_type}; use pipeline, vlm, or all")
|
||||
|
||||
print("Model download finished.")
|
||||
PY
|
||||
|
|
@ -0,0 +1,30 @@
|
|||
#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
|
||||
cd "$SCRIPT_DIR"
|
||||
|
||||
GPU_FLAVOR=${GPU_FLAVOR:-corex}
|
||||
|
||||
docker load -i unis-kb-latest.tar
|
||||
docker load -i unis-gateway-latest.tar
|
||||
docker load -i unis-mindmap-web-latest.tar
|
||||
|
||||
case "$GPU_FLAVOR" in
|
||||
corex)
|
||||
if [ -f unis-mindmap-api-corex-latest.tar ]; then
|
||||
docker load -i unis-mindmap-api-corex-latest.tar
|
||||
docker tag unis-mindmap-api:corex unis-mindmap-api:latest
|
||||
else
|
||||
docker load -i unis-mindmap-api-latest.tar
|
||||
fi
|
||||
;;
|
||||
cuda|nvidia)
|
||||
docker load -i unis-mindmap-api-cuda-latest.tar
|
||||
docker tag unis-mindmap-api:cuda unis-mindmap-api:latest
|
||||
;;
|
||||
*)
|
||||
echo "Unsupported GPU_FLAVOR: $GPU_FLAVOR. Use corex or cuda." >&2
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
|
|
@ -0,0 +1,40 @@
|
|||
#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
ARTIFACT_DIR=${ARTIFACT_DIR:-$(cd "$(dirname "$0")" && pwd)}
|
||||
cd "$ARTIFACT_DIR"
|
||||
|
||||
save_image() {
|
||||
local image="$1"
|
||||
local file="$2"
|
||||
echo "Saving ${image} -> ${file}"
|
||||
rm -f "$file"
|
||||
sudo docker save "$image" > "$file"
|
||||
ls -lh "$file"
|
||||
}
|
||||
|
||||
save_image_if_exists() {
|
||||
local image="$1"
|
||||
local file="$2"
|
||||
if sudo docker image inspect "$image" >/dev/null 2>&1; then
|
||||
save_image "$image" "$file"
|
||||
else
|
||||
echo "Skip ${image}: image not found"
|
||||
fi
|
||||
}
|
||||
|
||||
save_image unis-kb:latest unis-kb-latest.tar
|
||||
save_image unis-gateway:latest unis-gateway-latest.tar
|
||||
save_image unis-mindmap-api:latest unis-mindmap-api-latest.tar
|
||||
save_image_if_exists unis-mindmap-api:corex unis-mindmap-api-corex-latest.tar
|
||||
save_image_if_exists unis-mindmap-api:cuda unis-mindmap-api-cuda-latest.tar
|
||||
save_image unis-mindmap-web:latest unis-mindmap-web-latest.tar
|
||||
|
||||
checksum_files=(unis-*.tar)
|
||||
if compgen -G "mineru-models-*.tar.gz" > /dev/null; then
|
||||
checksum_files+=(mineru-models-*.tar.gz)
|
||||
fi
|
||||
checksum_files+=(*.sh README_DEPLOY.md DEPLOY_CN.md docker-compose.unis.yml)
|
||||
|
||||
sha256sum "${checksum_files[@]}" > SHA256SUMS.txt
|
||||
cat SHA256SUMS.txt
|
||||
|
|
@ -0,0 +1,15 @@
|
|||
#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
|
||||
cd "$SCRIPT_DIR"
|
||||
|
||||
export MINDMAP_PUBLIC_URL=${MINDMAP_PUBLIC_URL:-/mindmap/}
|
||||
export KB_DATA_DIR=${KB_DATA_DIR:-/opt/maxkb}
|
||||
export MINERU_MODEL_DIR=${MINERU_MODEL_DIR:-/data/unis/mineru-models}
|
||||
|
||||
sudo mkdir -p "$KB_DATA_DIR"
|
||||
sudo mkdir -p "$MINERU_MODEL_DIR"/pipeline "$MINERU_MODEL_DIR"/vlm
|
||||
sudo chmod -R 777 "$KB_DATA_DIR"
|
||||
sudo chmod -R 777 "$MINERU_MODEL_DIR"
|
||||
sudo -E docker compose -f docker-compose.unis.yml up -d
|
||||
|
|
@ -0,0 +1,20 @@
|
|||
#!/usr/bin/env bash
|
||||
set -euo pipefail
|
||||
|
||||
TARGET=${TARGET:?Set TARGET as user@host, for example root@10.100.52.80}
|
||||
TARGET_DIR=${TARGET_DIR:-/data1/zywlw/tar}
|
||||
ARTIFACT_DIR=${ARTIFACT_DIR:-$(cd "$(dirname "$0")" && pwd)}
|
||||
|
||||
ssh "$TARGET" "mkdir -p '$TARGET_DIR'"
|
||||
files=(
|
||||
"$ARTIFACT_DIR"/*.tar
|
||||
"$ARTIFACT_DIR"/*.sh
|
||||
"$ARTIFACT_DIR"/docker-compose.unis.yml
|
||||
"$ARTIFACT_DIR"/README_DEPLOY.md
|
||||
"$ARTIFACT_DIR"/SHA256SUMS.txt
|
||||
)
|
||||
if compgen -G "$ARTIFACT_DIR/mineru-models-*.tar.gz" > /dev/null; then
|
||||
files+=("$ARTIFACT_DIR"/mineru-models-*.tar.gz)
|
||||
fi
|
||||
|
||||
scp "${files[@]}" "$TARGET:$TARGET_DIR/"
|
||||
Loading…
Reference in New Issue