Hugging Face Spaces

Hugging Face is an open-source platform for machine learning and artificial intelligence that provides tools and models for building, training, and deploying AI applications. Hugging Face Spaces is a service that allows you to deploy machine learning models as web applications. Hugging Face Spaces can deploy Evidence apps from a Git repository.

Prerequisites

• A Hugging Face account
• An Evidence project pushed to a Git service like GitHub

Deploy Evidence to Hugging Face Spaces

1. Navigate to the Hugging Face Spaces

2. Select Create new Space

   • Choose a name
   • Space SDK: Static
   • Static template: Blank
   • Choose a visibility: Public or Private depending on your needs

3. Create a new Access Tokens

   • Token Type: Fine-grained
   • Name: e.g. Evidence
   • Repositories Permissions: Choose the space you just created, and select Read access to contents of selected repos and Write access to contents/settings of selected repos
   • Select Create Token, and save it somewhere safe

4. Add secrets to your GitHub repo: Settings > Secrets and variables > Actions

   • Firstly add your Hugging Face Tokent as a secret named HUGGINGFACE_TOKEN
   • With your Evidence dev server running, go to the settings page and copy each of the environment variables
   • Alternatively, you can find credentials in connection.options.yaml files in your /sources/your_source directory. The key format used should be EVIDENCE_SOURCE__[your_source]__[option_name] (Note the casing matches your source names, and the double underscores). Note that the values are base64 encoded, and will need to be decoded.

5. Create the deploy workflow file: .github/workflows/deploy.yml, update the repo name and space name, and merge it into your main branch

   name: Deploy to Hugging Face Space on Merge
   on:
   push:
      branches:
         - main
   workflow_dispatch:
   
   jobs:
   build_and_deploy:
      runs-on: ubuntu-latest
      steps:
         # Checkout the repository
         - uses: actions/checkout@v4
   
         # Install dependencies and build the project
         - run: npm ci && npm run sources && npm run build
         env:
            EVIDENCE_SOURCE__taxi__project_id: ${{ secrets.EVIDENCE_SOURCE__TAXI__PROJECT_ID }}
            EVIDENCE_SOURCE__taxi__client_email: ${{ secrets.EVIDENCE_SOURCE__TAXI__CLIENT_EMAIL }}
            EVIDENCE_SOURCE__taxi__private_key: ${{ secrets.EVIDENCE_SOURCE__TAXI__PRIVATE_KEY }}
   
         # Deploy to Hugging Face Space
         - name: Install Hugging Face CLI
         run: pip install huggingface-hub
   
         - name: Authenticate with Hugging Face
         run: huggingface-cli login --token "${{ secrets.HUGGINGFACE_TOKEN }}"
   
         - name: Deploy to Hugging Face Space
         run: |
            # Update to use your repo
            huggingface-cli upload [your-username]/[your-space-name] ./build --repo-type=space
   

Your app will be deployed to a URL like https://[your-username]-[your-space-name].static.hf.space/

Domains, Authentication and Scheduling

Authentication

If you chose a private space, only members of your Hugging Face organization will be able to access the space.

Custom Domains

Custom domains are not supported for Hugging Face Spaces.

Data refresh

You can adjust the schedule for your deployment using the workflow file by adding a schedule trigger with a cron expression.

on:
  push:
    branches: 'main'
  schedule:
    # This is every 10 minutes
    - cron: '*/10 * * * *'