# mineru-docker-cpu

**Repository Path**: MagX/mineru-docker-cpu

## Basic Information

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

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-08-18
- **Last Updated**: 2026-08-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# MinerU API Server - CPU-Only Docker Deployment

CPU-only deployment of [MinerU](https://github.com/opendatalab/MinerU) PDF/document parsing API server.

MinerU's official Docker deployment requires GPU (based on vllm). This provides a CPU-only alternative using the `pipeline` backend.

## Quick Start

### Option A: Use pre-built image (recommended)

```bash
# 1. Clone this repo (for docker-compose.yml)
git clone https://github.com/potterwhite/mineru-docker-cpu.git
cd mineru-docker-cpu

# 2. Pull the pre-built image
docker pull potterwhite/mineru-api-cpu:latest

# 3. Download pipeline models (~2.4GB)
docker compose run --rm mineru-api mineru-models-download -s modelscope -m pipeline

# 4. Start the server
docker compose up -d

# 5. Verify
curl http://localhost:8888/health
```

### Option B: Build from source

```bash
git clone https://github.com/potterwhite/mineru-docker-cpu.git
cd mineru-docker-cpu
docker compose build
docker compose run --rm mineru-api mineru-models-download -s modelscope -m pipeline
docker compose up -d
curl http://localhost:8888/health
```

## Usage

### Submit a document for parsing (async)

```bash
# Step 1: Submit task
TASK_ID=$(curl -s -X POST http://localhost:8888/tasks \
  -F "files=@document.pdf" \
  -F "backend=pipeline" | python3 -c "import sys,json; print(json.load(sys.stdin)['task_id'])")

# Step 2: Poll status
while true; do
  STATUS=$(curl -s http://localhost:8888/tasks/$TASK_ID | python3 -c "import sys,json; print(json.load(sys.stdin)['status'])")
  [[ "$STATUS" == "completed" || "$STATUS" == "failed" ]] && break
  sleep 10
done

# Step 3: Get result
python3 -c "
import urllib.request, json
data = json.loads(urllib.request.urlopen('http://localhost:8888/tasks/$TASK_ID/result', timeout=120).read())
for fname, content in data['results'].items():
    print(content['md_content'])
"
```

### API Endpoints

| Endpoint | Method | Description |
|---|---|---|
| `/health` | GET | Health check |
| `/tasks` | POST | Submit async parsing task (fields: `files`, `backend`) |
| `/tasks/{id}` | GET | Poll task status |
| `/tasks/{id}/result` | GET | Get full result JSON with `md_content` |

**Important**: Always pass `backend=pipeline` in POST requests. The default backend (`hybrid-auto-engine`) requires GPU.

## Configuration

### Models

Models are downloaded to `./models/` and persist across container restarts. The `mineru.json` config is auto-generated by `mineru-models-download` and mounted into the container.

- Pipeline models (CPU): ~2.4GB
- VLM models (GPU): larger, not needed for CPU-only

### Resource Limits

Default memory limit is 12GB. Adjust in `docker-compose.yml`:

```yaml
deploy:
  resources:
    limits:
      memory: 12G
```

### Concurrency

Default max concurrent requests: 3. Adjust via `--max-concurrent-requests` flag in the `command` field.

## Supported Formats

PDF, DOCX, PPTX, XLS, XLSX, images (PNG, JPG, WebP...)

## Performance

| Cores | 10-page PDF | 100-page PDF |
|---|---|---|
| 4 | ~30-60s | ~5-10min |
| 8 | ~15-30s | ~3-5min |
| 16 | ~10-15s | ~2-3min |

## Troubleshooting

```bash
# Check container status
docker compose ps

# View logs
docker compose logs -f

# Restart
docker compose restart

# Check if OOM killed
docker inspect mineru-api | grep -i oom
```

## License

MinerU is licensed under the [MinerU Open Source License](https://github.com/opendatalab/MinerU) (based on Apache 2.0).