kn1ght

A work in progress...

Tokenizer Differences

kn1ght's tokenizer is optimized for Chess' Portable Game Notation (PGN) format.

Note: kn1ght's tokenizer does not currently account for PGN metadata (Event, Site, Date, etc.), PGN comments ({...}), notes about clock times ({[%clk ...]}), or other miscellaneous PGN data. It only focuses on the actual moves played in the game.

It has been trained on a small dataset of 3.5M chess games from ChessDB cleaned up by Kaggle user milesh1.