runpod-instructions.md

Deploying Marker on RunPod

Prerequisites

• A running RunPod instance with GPU support
• Access to the Web Terminal or SSH

Steps to Deploy

1. Access your RunPod instance

   • Go to your RunPod dashboard
   • Find your pod and click on "Connect"
   • In the Connection Options dialog, click "Start" under Web Terminal
   • This will open a terminal in your browser

2. Clone the repository

   • In the Web Terminal, run:

   cd /workspace
   git clone https://github.com/yourusername/marker.git
   cd marker

   • Alternatively, you can upload your local files using the RunPod file browser

3. Build the Docker image

   • In the Web Terminal, run:

   docker build -t marker-gpu-api -f Dockerfile.gpu .

4. Run the container with GPU support

   • In the Web Terminal, run:

   docker run -d --gpus all -p 8080:8080 \
     -e PORT=8080 \
     marker-gpu-api

5. Expose the API endpoint

   • Go back to the RunPod dashboard
   • Click on "Connect" for your pod
   • In the Connection Options dialog, under "HTTP Services"
   • Click "Add HTTP Service"
   • Enter:
     • Internal Port: 8080
     • Service Name: Marker API
   • Click "Add"

6. Access the API

   • Once the service is added, you'll see a URL in the HTTP Services section
   • Click on the URL to access your Marker API

Troubleshooting

If you encounter any issues:

1. Check container logs

   docker ps  # Get the container ID
   docker logs <container_id>

2. Check GPU availability

   nvidia-smi

3. Verify port mapping

   docker ps

   Ensure the container is properly mapping port 8080