metadata
library_name: transformers
tags:
- chess
license: mit
Chess GPT - Prof's Architecture
Params: 998,656 Vocab: 1604 (TOP_K=2000) Dataset: 1M samples x 5 epochs
Config:
- n_embd: 128
- n_layer: 4
- n_head: 4
- LR: 5e-4
- UNK rate: 25.7%
Target: 60-70% legal rate