# Ares Starter Dataset

This folder contains public starter data for the Ares Static Lab UI.

## Files

- `ares_roleplay_starter_1000.jsonl` — 1,000 deterministic roleplay chat scenarios.
- `ares_ml_process_starter_500.txt` — 500 deterministic machine-learning process notes and chat examples.

## Larger dataset generation

The full Colab notebook generates a larger training mixture on demand:

- Wikipedia streamed from `wikimedia/wikipedia`
- ML-process curriculum
- 12,000+ roleplay scenarios by default
- train/validation splits
- tokenizer training data
- optional roleplay SFT data

Notebook:

```text
colab/Ares_Wikipedia_Roleplay_Brain_Training.ipynb
```

## Why the full Wikipedia dataset is not committed here

Wikipedia can be very large. The notebook streams bounded samples from Hugging Face instead of committing large dumps into this Static Space repo.

## License / usage notes

- Synthetic Ares curriculum files in this repository are project-generated starter data.
- Wikipedia records are loaded from the Hugging Face `wikimedia/wikipedia` dataset in Colab; follow the dataset's licensing and attribution requirements.
- Do not add private or sensitive data to public training corpora.