/chess_llm_interpretability

Visualizing the internal board state of a GPT trained on chess PGN strings, and performing interventions on its internal board state and representation of player Elo.

Primary LanguageJupyter Notebook