forked from qoherent/modrec-workflow
23 lines
793 B
Markdown
23 lines
793 B
Markdown
|
# RIA Hub Technical Demo
|
||
|
|
|
||
|
|
This repository demonstrates a full ML pipeline via Gitea Actions:
|
||
|
|
|
||
|
|
- **Recordings**
|
||
|
|
A collection of raw `.npy` radio recordings stored via Git LFS.
|
||
|
|
|
||
|
|
- **Workflows**
|
||
|
|
A CI pipeline that automatically:
|
||
|
|
1. Builds a labeled dataset from raw recordings
|
||
|
|
2. Trains a model on that dataset
|
||
|
|
3. Optimizes the model and packages an inference application
|
||
|
|
|
||
|
|
- **Scripts**
|
||
|
|
- `scripts/build_dataset.sh`
|
||
|
|
Reads through `recordings/`, applies preprocessing, and outputs training `.npz` or `.csv` files into `data/`.
|
||
|
|
- `scripts/train_model.sh`
|
||
|
|
Consumes `data/`, trains a PyTorch model, and writes checkpoints to `checkpoints/`.
|
||
|
|
- `scripts/build_app.sh`
|
||
|
|
Takes the best checkpoint and builds a small inference CLI or server in `dist/`.
|
||
|
|
|
||
|
|
|
||
|
|
|