chess-bot-3000-chess960-ft

Fine-tuned version of daavidhauser/chess-bot-3000-100m by David Hauser — a 106M-parameter Qwen2 causal language model pre-trained on ~100M Lichess games.

This model was fine-tuned on Chess960 (Fischer Random Chess) games for 3 epochs as part of an experiment studying memorisation vs. principled play in chess language models (ECS7026P, Queen Mary University of London, 2026).

Load the model

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("nyuzbashev/chess-bot-3000-chess960-ft")
tokenizer = AutoTokenizer.from_pretrained("nyuzbashev/chess-bot-3000-chess960-ft")

Credits

Original model: daavidhauser/chess-bot-3000-100m