Skip to content

cgpu-setup.md

Latest commit

 

History

History
77 lines (58 loc) · 2.76 KB

cgpu-setup.md

File metadata and controls

77 lines (58 loc) · 2.76 KB

cgpu Setup Guide (Custom Credentials)

The default cgpu configuration uses a shared developer project which is currently restricted. To use cgpu correctly, you need to set up your own Google Cloud credentials.

Step 1: Create a Google Cloud Project

  1. Go to the Google Cloud Console.
  2. Click the project selector (top left) and choose New Project.
  3. Name it montage-ai-cgpu (or anything you like) and click Create.

Step 2: Configure OAuth Consent Screen

  1. Go to APIs & Services > OAuth consent screen.
  2. Choose External (available to any user with a Google account) and click Create.
    • Note: If you choose Internal, only users in your organization can use it.
  3. App Information:
    • App name: cgpu
    • User support email: Your email
    • Developer contact information: Your email
  4. Click Save and Continue.
  5. Scopes: Click Add or Remove Scopes.
    • Manually add this scope: https://www.googleapis.com/auth/colaboratory
    • Also select .../auth/userinfo.email and .../auth/userinfo.profile.
    • Click Update, then Save and Continue.
  6. Test Users:
    • Click Add Users.
    • Enter your own Google email address.
    • Click Save and Continue.

Step 3: Create Credentials

  1. Go to APIs & Services > Credentials.
  2. Click Create Credentials > OAuth client ID.
  3. Application type: Select Desktop app.
  4. Name: cgpu-client
  5. Click Create.
  6. Copy the Client ID and Client Secret.

Step 4: Configure cgpu

Provide the Client ID and Client Secret to the assistant, or create the config file manually at ~/.config/cgpu/config.json:

{
  "clientId": "YOUR_CLIENT_ID",
  "clientSecret": "YOUR_CLIENT_SECRET",
  "colabApiDomain": "https://colab.research.google.com",
  "colabGapiDomain": "https://colab.googleapis.com"
}

OAuth Login (Interactive)

Run the login locally (not inside the cluster):

cgpu connect

Notes:

  • Keep the terminal open until the flow completes.
  • If the callback page shows “connection refused”, the local callback server wasn’t running. Re-run cgpu connect and retry the login.

Cluster / Kubernetes

If you run Montage AI in the cluster, CGPU jobs (encode/upscale/voice isolation) need the cgpu config mounted into worker pods. Create a secret named cgpu-credentials with config.json and optional session.json, then redeploy: the workers mount it into /home/montage/.config/cgpu. Without this secret, CGPU jobs will fall back to local CPU/GPU.

One‑shot helper (recommended)

./scripts/ops/cgpu-refresh-session.sh

This runs cgpu connect, updates the secret, and restarts the worker + cgpu-server deployments.