---
license: apache-2.0
datasets:
- open-chess/MetaChess-20k
language:
- en
base_model:
- Qwen/Qwen2.5-7B-Instruct
pipeline_tag: text-generation
tags:
- chess
- reasoning
---

# Caissa-Chess-M1

**Caissa-Chess-M1** is a chess reasoning model that thinks like a grandmaster. Based on Qwen 2.5 7B, fine-tuned with LoRA on 20,000 complex positions with detailed Chain-of-Thought explanations.

## 🎯 Features

- **Structured reasoning** with `<evaluation>`, `<calculation>`, and `<reasoning>` tags
- **Grandmaster-level analysis** trained on Stockfish depth 28+ positions
- **Chain-of-Thought** explanations for every move
- **Balanced positions** (-400 to +400 centipawns)

---

## 📖 Citation

```bibtex
@misc{metachess20k,
  author = {Open Chess},
  title = {MetaChess-20k: Chess Reasoning Dataset with Chain-of-Thought},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/open-chess/MetaChess-20k}
}
```

---