# Modal — LLM polish workers

Deploy three GPU workers (MiniCPM, Aya, Nemotron) for parallel polish on the hosted Space.

```bash
.venv/bin/modal deploy modal_llm.py -e main
```

**App name:** `orions-cottage-llm`

### HF Space secrets

| Secret | Value |
|--------|--------|
| `MODAL_TOKEN_ID` | From the workspace that ran `modal deploy` |
| `MODAL_TOKEN_SECRET` | Pairing secret |
| `MODAL_ENVIRONMENT` | `main` |
| `MODAL_APP_NAME` | `orions-cottage-llm` |

Dashboard: `https://modal.com/apps/<workspace>/main/deployed/orions-cottage-llm`

First Plan after idle may take 1–2 minutes (cold GPU + model pull).