# manak0/Detect-fire

Specialized **public-track** package for Score Vision SN44 WTA element
`manak0/Detect-fire` (weight 0.04, allocation=wta).

| Field | Value |
|-------|-------|
| HF repo target | `synapseml/sn44-detect-fire` |
| Short name | `fire` |
| Max package size | **≤ 30 MB** (`CHECKER_MAX_MODEL_BYTES=31457280`) |
| Track | public / open-source (HF + Chutes) |

## Object labels (specialist)

`fire`, `smoke`, `flame`, `ember`, `torch`

## Required files

- `miner.py` — chute entrypoint (`Miner.predict_batch`)
- `chute_config.yml` — Chutes runtime/image
- `element.json` — packaging metadata for control plane
- `labels.json` — object list for distill / bench

## Size gate

Uploaded HF revision snapshot must stay under
`31457280` bytes. Measured post-export via:

```bash
uv run python -m sn44_elite.public.size_gate --path . --element-id manak0/Detect-fire
```

## Deploy dry-run

```bash
bash scripts/deploy_os_dry_run.sh --element-id manak0/Detect-fire --model-path .
```

This scaffold does **not** force a mono multi-element blob: each public
element lives in its own package directory and HF repo.