# Setup Modal

> Deploy the CLaaS Modal serverless distillation API. Use when deploying the SDPO distillation service to production.

- Skill: `experientiallabs/setup-modal` (Agent Skill)
- Install (CLI): `npx skillmds@latest add experientiallabs/setup-modal`
- Raw SKILL.md: https://api.skillmd.com/api/skills/experientiallabs/setup-modal/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: experientiallabs (https://skillmd.com/u/experientiallabs)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/experientiallabs/setup-modal

---


# 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**:
   ```bash
   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:

```bash
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