Fix large config: correct sliding window + num_iterations/pct_trained (runs used medium x2 iters, not iteration_counts/) 473c627 verified mugezhang commited on 25 days ago
Slim checkpoints to model-only (drop step/val_loss/code); update README/loader b9f2e1a verified mugezhang commited on Jul 5
Add 9 multilingual value-embedding GPT models (text/romanized/ipa x small/medium/large) + loader a5745bb verified mugezhang commited on Jul 5