Ray HPO Example with GPU Support

Environment Variables

STORAGE_PATH=s3://your-bucket-name/ray-results
AWS_ACCESS_KEY_ID=your_access_key_id
AWS_SECRET_ACCESS_KEY=your_secret_access_key
AWS_DEFAULT_REGION=us-east-1
AWS_ENDPOINT_URL=

Required Environment Variables:

• STORAGE_PATH - S3 path for storing Ray checkpoints (format: s3://bucket-name/path)
• AWS_ACCESS_KEY_ID - AWS access key for S3 authentication
• AWS_SECRET_ACCESS_KEY - AWS secret key for S3 authentication

Optional Environment Variables:

• AWS_DEFAULT_REGION - AWS region (default: us-east-1)
• AWS_ENDPOINT_URL - Custom S3 endpoint (for S3-compatible storage like MinIO, leave empty for AWS S3)

Environment Variables

Worker Node

• RAY_HEAD_ADDRESS (required): Address of the Ray head node (e.g., ray-head:6379)

Ports

Head Node

• 6379: Ray GCS (Global Control Service)
• 8265: Ray Dashboard (web UI)
• 10001: Ray Client Server

Worker Nodes

• No exposed ports required (connects to head node)

Running the Job

Connect to the Head node and use the Ray CLI to submit the Job:

ray job submit --working-dir . -- uv run hpo_example.py