mirror of
https://github.com/opendatalab/MinerU.git
synced 2026-09-19 09:30:53 +08:00
feat: add --enable-vlm-preload option to CLI for VLM model preloading during startup
This commit is contained in:
myhloli
@@ -34,6 +34,8 @@ Options:
|
||||
--host TEXT Server host (default: 127.0.0.1)
|
||||
--port INTEGER Server port (default: 8000)
|
||||
--reload Enable auto-reload (development mode)
|
||||
--enable-vlm-preload BOOLEAN
|
||||
Preload the local VLM model during mineru-api startup.
|
||||
--help Show this message and exit.
|
||||
```
|
||||
```bash
|
||||
@@ -55,6 +57,8 @@ Options:
|
||||
--server-port INTEGER Set the server port for the Gradio app.
|
||||
--api-url TEXT MinerU FastAPI base URL. If omitted, gradio
|
||||
starts a reusable local mineru-api service.
|
||||
--enable-vlm-preload BOOLEAN Preload the local VLM model when gradio
|
||||
starts a local mineru-api service.
|
||||
--latex-delimiters-type [a|b|all]
|
||||
Set the type of LaTeX delimiters to use in
|
||||
Markdown rendering: 'a' for type '$', 'b' for
|
||||
@@ -73,6 +77,9 @@ Options:
|
||||
--local-gpus TEXT Local GPU workers to launch: auto, none, or CSV such
|
||||
as 0,1,2
|
||||
--worker-host TEXT Host for router-managed workers (default: 127.0.0.1)
|
||||
--enable-vlm-preload BOOLEAN
|
||||
Preload the local VLM model in router-managed
|
||||
mineru-api workers.
|
||||
--help Show this message and exit.
|
||||
```
|
||||
|
||||
@@ -134,6 +141,11 @@ Here are the environment variables and their descriptions:
|
||||
* Used to configure the root output directory for `mineru-api`.
|
||||
* Default is `./output` under the current working directory.
|
||||
|
||||
- `MINERU_LOCAL_API_STARTUP_TIMEOUT_SECONDS`:
|
||||
* Used to control how long CLI tools wait for a locally started `mineru-api` to become healthy.
|
||||
* Default is `300` seconds.
|
||||
* Applies to temporary local API startup in `mineru`, preload startup in `mineru-gradio`, and router-managed local workers.
|
||||
|
||||
- `MINERU_API_TASK_RETENTION_SECONDS`:
|
||||
* Used to set how long completed or failed tasks are retained, in seconds.
|
||||
* Default is `86400` seconds (24 hours).
|
||||
|
||||
@@ -48,6 +48,7 @@ If you need to adjust parsing options through custom parameters, you can also ch
|
||||
>Tasks are tracked only in-process for a single `mineru-api` instance. Task status is not preserved across service restarts, `--reload`, or multi-process deployments.
|
||||
>Completed or failed tasks are retained for 24 hours by default, then their task state and output directory are cleaned automatically. After cleanup, task status and result endpoints return `404`.
|
||||
>Use `MINERU_API_TASK_RETENTION_SECONDS` and `MINERU_API_TASK_CLEANUP_INTERVAL_SECONDS` to adjust retention and cleanup polling intervals.
|
||||
>Use `--enable-vlm-preload true` to warm up the local VLM model during service startup instead of waiting for the first VLM or hybrid request.
|
||||
>
|
||||
>Asynchronous task submission example:
|
||||
>```bash
|
||||
@@ -82,6 +83,7 @@ If you need to adjust parsing options through custom parameters, you can also ch
|
||||
>
|
||||
>- Access `http://127.0.0.1:7860` in your browser to use the Gradio WebUI.
|
||||
>- Without `--api-url`, Gradio starts a reusable local `mineru-api`; with `--api-url`, it reuses an existing local or remote service.
|
||||
>- `--enable-vlm-preload true` makes Gradio start its local `mineru-api` during WebUI startup and wait for VLM preload to finish. It is ignored when `--api-url` points to an existing service.
|
||||
>- The WebUI currently accepts `PDF`, image, and `DOCX` uploads.
|
||||
|
||||
- Use `mineru-router` for multi-service / multi-GPU orchestration:
|
||||
@@ -92,6 +94,7 @@ If you need to adjust parsing options through custom parameters, you can also ch
|
||||
>
|
||||
>- `mineru-router` exposes the same `/health`, `/tasks`, `/file_parse`, `/tasks/{task_id}`, and `/tasks/{task_id}/result` interface set as `mineru-api`.
|
||||
>- Repeat `--upstream-url` to aggregate multiple existing `mineru-api` services, or use `--local-gpus` to launch local workers automatically.
|
||||
>- `--enable-vlm-preload true` only applies to router-managed local workers. It does not preload remote services passed through `--upstream-url`.
|
||||
>- It is intended for advanced multi-service, multi-GPU, and unified-entry deployments.
|
||||
|
||||
- Using `http-client/server` method:
|
||||
|
||||
@@ -34,6 +34,8 @@ Options:
|
||||
--host TEXT 服务器主机地址（默认：127.0.0.1）
|
||||
--port INTEGER 服务器端口（默认：8000）
|
||||
--reload 启用自动重载（开发模式）
|
||||
--enable-vlm-preload BOOLEAN
|
||||
在 mineru-api 启动阶段预加载本地 VLM 模型
|
||||
--help 显示此帮助信息并退出
|
||||
```
|
||||
```bash
|
||||
@@ -50,6 +52,8 @@ Options:
|
||||
--server-port INTEGER 设置 Gradio 应用程序的服务器端口
|
||||
--api-url TEXT MinerU FastAPI 服务地址；不传时自动拉起可复用的本地
|
||||
mineru-api
|
||||
--enable-vlm-preload BOOLEAN 在 Gradio 拉起本地 mineru-api 时预加载本地
|
||||
VLM 模型
|
||||
--latex-delimiters-type [a|b|all]
|
||||
设置在 Markdown 渲染中使用的 LaTeX 分隔符类型
|
||||
('a' 表示 '$' 类型，'b' 表示 '()[]' 类型，
|
||||
@@ -67,6 +71,9 @@ Options:
|
||||
--upstream-url TEXT 现有 MinerU FastAPI 服务地址；可重复传入多个
|
||||
--local-gpus TEXT 本地 GPU worker 配置：auto、none 或 0,1,2 形式
|
||||
--worker-host TEXT 路由托管 worker 的监听地址（默认：127.0.0.1）
|
||||
--enable-vlm-preload BOOLEAN
|
||||
在 router 托管的本地 mineru-api worker 中预加载本地
|
||||
VLM 模型
|
||||
--help 显示此帮助信息并退出
|
||||
```
|
||||
|
||||
@@ -128,6 +135,11 @@ MinerU命令行工具的某些参数存在相同功能的环境变量配置，
|
||||
* 用于指定 `mineru-api` 输出目录根路径
|
||||
* 默认为当前工作目录下的 `./output`。
|
||||
|
||||
- `MINERU_LOCAL_API_STARTUP_TIMEOUT_SECONDS`：
|
||||
* 用于控制各命令行工具等待本地拉起的 `mineru-api` 进入健康状态的最长时间
|
||||
* 默认为 `300` 秒。
|
||||
* 适用于 `mineru` 的临时本地 API、`mineru-gradio` 的 preload 启动，以及 `mineru-router` 托管的本地 worker。
|
||||
|
||||
- `MINERU_API_TASK_RETENTION_SECONDS`：
|
||||
* 用于设置任务完成或失败后的保留时长（秒）
|
||||
* 默认为 `86400` 秒（24 小时）。
|
||||
|
||||
@@ -48,6 +48,7 @@ mineru -p <input_path> -o <output_path>
|
||||
>任务为单进程、进程内状态实现，服务重启、`--reload` 热重载或多进程部署后不保证仍可查询历史任务状态。
|
||||
>默认任务完成或失败后保留 24 小时，随后自动清理任务状态和输出目录；清理后访问任务状态或结果会返回 `404`。
|
||||
>可通过环境变量 `MINERU_API_TASK_RETENTION_SECONDS` 和 `MINERU_API_TASK_CLEANUP_INTERVAL_SECONDS` 调整保留时长与清理轮询间隔。
|
||||
>可通过 `--enable-vlm-preload true` 在服务启动阶段预热本地 VLM 模型，避免首次 VLM 或 hybrid 请求时再初始化。
|
||||
>
|
||||
>异步任务提交示例：
|
||||
>```bash
|
||||
@@ -82,6 +83,7 @@ mineru -p <input_path> -o <output_path>
|
||||
>
|
||||
>- 在浏览器中访问 `http://127.0.0.1:7860` 使用 Gradio WebUI。
|
||||
>- 未传 `--api-url` 时，Gradio 会自动拉起可复用的本地 `mineru-api`；传入 `--api-url` 时则会复用已有本地或远端服务。
|
||||
>- `--enable-vlm-preload true` 会让 Gradio 在 WebUI 启动阶段主动拉起本地 `mineru-api` 并等待 VLM 预加载完成；传入 `--api-url` 时会被忽略。
|
||||
>- WebUI 当前支持上传 `PDF`、图片与 `DOCX` 文件。
|
||||
|
||||
- 通过 `mineru-router` 进行多服务 / 多 GPU 编排：
|
||||
@@ -92,6 +94,7 @@ mineru -p <input_path> -o <output_path>
|
||||
>
|
||||
>- `mineru-router` 对外暴露与 `mineru-api` 一致的 `/health`、`/tasks`、`/file_parse`、`/tasks/{task_id}`、`/tasks/{task_id}/result` 接口。
|
||||
>- 可重复使用 `--upstream-url` 聚合多个已有 `mineru-api` 服务，也可通过 `--local-gpus` 自动拉起本地 worker。
|
||||
>- `--enable-vlm-preload true` 仅作用于 router 托管的本地 worker，不会影响通过 `--upstream-url` 接入的远端服务。
|
||||
>- 适用于多服务、多 GPU 和统一入口部署场景。
|
||||
|
||||
- 使用`http-client/server`方式调用：
|
||||
|
||||
Reference in New Issue
Block a user