Skip to main content
Version: 0.4.3

Working with GCP

Constellaxion integrates seamlessly with Google Cloud Platform (GCP) to deploy and fine-tune open-source models via Vertex AI and Cloud Storage.

This page explains what resources the CLI sets up behind the scenes and how to prepare your project.

What Constellaxion Does on GCP​

When you run a commands like constellaxion init, or constellaxion model deploy, Constellaxion performs a series of actions behind the scenes to enable the right GCP resources to handle the job. Some examples of what Constellaxion might do:

  • Create or launch Vertex AI custom training jobs
  • Upload datasets and model artifacts to Cloud Storage
  • Deploy models to Vertex AI Endpoints
  • Manage job metadata using service accounts

These actions require specific IAM permissions on your GCP project.

✅ Required Roles​

Your Google Cloud user account must be assigned:

  • roles/resourcemanager.projectIamAdmin – to create and manage service accounts

👉 Granting IAM roles to a user

⚙️ Enabling Vertex AI​

You must enable the Vertex AI API in your GCP project:

👉 Enable the Vertex AI API

🛡️ Service Account Setup​

After you run constellaxion init, Constellaxion creates a service account named:

constellaxion-admin@{project_id}.iam.gserviceaccount.com

This account is used to run training and deployment jobs on Vertex AI.

It will be granted the following roles:

PolicyPurpose
roles/resourcemanager.projectIamAdminTo create and manage service accounts
roles/aiplatform.userTo create and manage Vertex AI resources
roles/storage.adminTo read/write training data and model artifacts
Security Note

This service account approach is not ideal for production deployments. We're working on a more secure solution. If you have any urgent needs, please reach out to us at support@constellaxion.ai.

🌍 Region Support​

Constellaxion supports all regions where Vertex AI is available. You configure this in your model.yaml file under the deploy.gcp.location field.

deploy:
  gcp:
    project_id: my-project
    region: us-central1

👉 Available Vertex AI regions

🧪 Next Steps​

Once GCP is set up:

  1. Create your model directory with a model.yaml file
  2. Run constellaxion init
  3. You’re ready to deploy, fine-tune, and prompt your models 🚀