# marker-example

**Repository Path**: felix-hua/marker-example

## Basic Information

- **Project Name**: marker-example
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-04-12
- **Last Updated**: 2026-04-12

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# file-convert-service

Standalone FastAPI service for PDF-to-Markdown conversion.

## What This Repo Contains

- A self-contained HTTP service for converting uploaded PDF files.
- Local development support with `uvicorn`.
- Optional containerized startup with `docker compose`.
- Persistent model caches for marker-related downloads when running with Docker Compose.

## Local Development

Install runtime dependencies:

```powershell
python -m pip install -r requirements.txt
```

Install test dependencies:

```powershell
python -m pip install -r requirements-dev.txt
```

Start the API locally:

```powershell
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000
```

Run the test suite:

```powershell
python -m pytest -q
```

For `docker compose`, create a local `.env` from the example file if you want to override the default port, log settings, or cache paths. For local `uvicorn` runs, set environment variables in your shell as needed.

## Docker

Build the image from this repository root:

```powershell
docker build -t file-convert-service .
```

The Dockerfile uses BuildKit pip cache mounts, so repeated builds can reuse downloaded Python packages instead of pulling them every time.

Start the service with Docker Compose:

```powershell
docker compose up --build
```

The API will be available on `http://localhost:8000`.

## Configuration

The service reads these environment variables:

- `APP_ENV` default `development`
- `APP_LOG_LEVEL` default `INFO`
- `APP_LOG_FORMAT` default `auto`
- `APP_LOG_ACCESS_ENABLED` default `true`
- `HF_HOME` default `/data/huggingface`
- `MODEL_CACHE_DIR` default `/data/datalab/models`

Example values are provided in `.env.example`.

## Model Cache Behavior

- `docker compose` mounts named volumes for Hugging Face and Datalab model caches.
- The first conversion request may still download marker-related models.
- After the first download, subsequent container restarts and rebuilds reuse the persisted cache and should not download the same model files again.

## Dependencies

- `marker-pdf==1.10.2` is declared explicitly in `requirements.txt`.

## API

- `GET /health`
  - Returns `{"status": "ok"}`
- `POST /internal/converters/pdf-to-markdown/file`
  - Converts an uploaded PDF and returns Markdown plus inline extracted images
  - Request type: `multipart/form-data`
  - Form fields: `file=<uploaded .pdf>`, optional `model=marker`

## Notes

- Request ID compatibility is preserved with both `X-Request-ID` and `X-Convert-Task-Id`.
- The service no longer depends on object storage and only supports direct file uploads.
- If Docker BuildKit is disabled locally, the pip cache mount in the Dockerfile will not be used.