open-chess/Caissa-Chess-M1
Caissa-Chess-M1 is a 7.6 billion parameter chess reasoning model developed by Open Chess, based on the Qwen 2.5 architecture. It is fine-tuned with LoRA on 20,000 complex chess positions, providing grandmaster-level analysis and Chain-of-Thought explanations. This model specializes in structured reasoning for chess, utilizing tags for evaluation, calculation, and reasoning to explain moves.
Loading preview...
Caissa-Chess-M1: Grandmaster Chess Reasoning
Caissa-Chess-M1 is a specialized 7.6 billion parameter model from Open Chess, built upon the Qwen 2.5 architecture. It is uniquely fine-tuned using LoRA on a dataset of 20,000 complex chess positions, each accompanied by detailed Chain-of-Thought explanations. This training methodology enables the model to perform grandmaster-level analysis, focusing on positions with balanced centipawn values (-400 to +400).
Key Capabilities
- Structured Reasoning: Provides analysis using distinct
<evaluation>,<calculation>, and<reasoning>tags. - High-Level Analysis: Trained on positions analyzed to Stockfish depth 28+, ensuring sophisticated chess insights.
- Transparent Explanations: Offers Chain-of-Thought explanations for every suggested move, detailing the rationale.
Good For
- Chess Analysis Tools: Integrating advanced chess reasoning and explanation capabilities.
- Educational Platforms: Demonstrating complex chess strategies and thought processes.
- AI Chess Research: Exploring structured reasoning and Chain-of-Thought in domain-specific applications.