| name | upload-model-weights |
| description | Upload ML model weights to GitHub Releases. Use when managing model files, re-uploading weights, or setting up model hosting on GitHub Releases. |
| allowed-tools | Bash, Read |
| disable-model-invocation | true |
Upload Model Weights
Pipeline
HuggingFace Hub (gated) → Download → Rename with prefix → GitHub Release (tag: models-v1.0)
Current Model
| Model | HF Repo | Release Tag | Files |
|---|
| Small (341M) | stabilityai/stable-audio-open-small | models-v1.0 | stable-audio-open-small-model.safetensors, stable-audio-open-small-model_config.json |
Triggering (Manual Only)
- Go to: https://github.com/mpolatcan/ccbell-sound-generator/actions/workflows/upload-model-weights.yml
- Click "Run workflow" → set tag (default:
models-v1.0)
The workflow downloads from HF Hub, renames with model prefix, and uploads to GitHub Release.
Required Secrets
| Secret | Description |
|---|
HF_TOKEN | HuggingFace token (read access to gated models) |
How the App Downloads Models
- Primary: GitHub Releases (
CCBELL_MODEL_DOWNLOAD_BASE_URL)
- Fallback: HuggingFace Hub (if GitHub fails AND
HF_TOKEN is set)
- Cache:
~/.cache/ccbell-models/stable-audio-open-{model_id}/
Verifying
gh release view models-v1.0 --json assets --jq '.assets[].name'
