setup-modal

name	setup-modal
description	Deploy the CLaaS Modal serverless distillation API. Use when deploying the SDPO distillation service to production.

Deploy Modal Worker

Deploy the Modal serverless distillation API for CLaaS.

Instructions

When this skill is invoked, perform the following steps:

1. Deploy the Modal app:

   modal deploy -m claas.modal.deploy
   

2. Extract the endpoint URL from the deployment output (look for the line with the FastAPI endpoint URL)

3. Report the deployment status including:

   • The deployed endpoint URL
   • Link to the Modal dashboard for the deployment
   • Any warnings or errors from the deployment

Deployed Services

The deployment includes:

• DistillWorker: L40S GPU worker with Qwen3-8B student model (self-distillation)
• FastAPI endpoint: REST API for feedback and distillation requests

Health Check

After deployment, verify the services are healthy:

curl https://{your-modal-app-url}/v1/health

Notes

• The Modal token must already be configured (run modal token set if not)
• First deployment takes longer due to model downloads (~5-10 minutes)
• Subsequent deployments are faster if the images haven't changed
• GPU memory snapshots enable sub-second cold starts after initial warmup