Create README.md
Browse files---
language: en
tags:
- chess
- language-model
- leon-llm
license: mit
---
# Chess Language Model
Explore the model's capabilities and interact with it on [GitHub](https://github.zhaw.ch/schmila7/leon-llm).
This model is trained on chess game sequences using a custom notation, xLAN, and aims to predict legal chess moves and understand chess game dynamics. The model is an adaptation of GPT-2 trained from scratch, tailored to comprehend and generate chess moves.
## Model description
Our Chess Language Model is a transformer-based model trained on a large corpus of chess games. The training process involved adapting the GPT-2 architecture to understand and predict chess moves in our custom xLAN format. xLAN helps in providing a uniform and fixed-length format for each move, enhancing the model's ability to learn the intricacies of chess.
**Training and Evaluation Parameters:**
(Insert specific training and evaluation parameters here.)
## Intended uses & limitations
The model is designed for:
- Generating legal chess moves.
- Analyzing chess positions.
- Predicting outcomes of chess games.
### How to use
Use our Notebooks on [GitHub](https://github.zhaw.ch/schmila7/leon-llm) to use it, because it needs a custom tokenizer to be used.
## Training Data
The Chess Language Model was trained on a comprehensive dataset sourced from the [Lichess database](https://database.lichess.org/), specifically using the collection from September 2023 with 93M games.
## Evaluation results
