GitHub Actions Self-Hosted GPU Runner

GitHub Actions hosted runners do not have GPUs. If your CI tests GPU code, runs small fine-tunes, or validates inference changes, you need a self-hosted runner. On our dedicated GPU hosting standing one up takes 15 minutes.

Contents

• Setup
• Workflow using the runner
• Isolation
• Security

Setup

In your GitHub repo settings, go to Actions -> Runners -> New self-hosted runner. Follow the displayed commands on your GPU server:

mkdir actions-runner && cd actions-runner
curl -o actions-runner-linux-x64-2.321.0.tar.gz -L \
  https://github.com/actions/runner/releases/download/v2.321.0/actions-runner-linux-x64-2.321.0.tar.gz
tar xzf actions-runner-linux-x64-2.321.0.tar.gz
./config.sh --url https://github.com/yourorg/yourrepo --token <token>
./run.sh

Install as a service:

sudo ./svc.sh install
sudo ./svc.sh start

Workflow

name: GPU Tests
on: [push]
jobs:
  test:
    runs-on: self-hosted
    steps:
      - uses: actions/checkout@v4
      - run: nvidia-smi
      - run: pip install -r requirements.txt
      - run: pytest tests/gpu/

Target the runner via runs-on: self-hosted or use custom labels if you have multiple runners with different GPUs.

Isolation

GitHub runners run arbitrary code from your CI pipeline. Do not share a runner with your production inference. Best practices:

• Dedicated user account with minimal permissions
• Ephemeral runners for each job (--ephemeral flag)
• Run in a container with GPU passthrough to isolate filesystem
• No production secrets on the runner host

Security

Do not enable self-hosted runners on public repos – any PR from a stranger could run arbitrary code on your GPU. For public repos use GitHub’s cloud runners or require manual approval.

See remote VS Code.