# ICC Demo Workspace

Recordings, curated datasets, and training workflows for the WavesFM ICC presentation.

## Workflow

1. **Upload recordings** — drop `.sigmf-data` + `.sigmf-meta` pairs. Git LFS tracks them automatically via `.gitattributes`.
2. **Curate** — open the Curator UI, select recordings from this repo, configure slicer + qualifier, produce an HDF5 dataset.
3. **Commit dataset** — use the Curator's "Commit to Repository" button to land the curated `.h5` back here.
4. **Train** — open the Model Trainer, select this repo + WavesFM Linear Probe (or LoRA), pick the dataset, submit the run.
5. **Watch action_run** — the trainer renders `.riahub/workflows/train.yaml` and triggers a runner job. Progress lives under the Actions tab.

## Notes

- WavesFM foundation model: `qoherent/wavesfm-base/wavesfm-v1p0.pth` — injected into the workflow automatically.
- If you upload a binary format not in `.gitattributes`, add it BEFORE the first commit of that file (LFS can't retroactively un-track).