nikitastheo/BERTtime-Stories-10m-nucleus-1-balanced

This model uses the LTG-BERT architecture. The model was trained on a combination of the BabyLM Dataset, the TinyStories Dataset, and generated data, in accordance with the rules of the Stric-Small track, and the 10M word budget.

The model was trained with 128 token sequence length

Hyperparameters used and evaluation scores will follow in a subsequent update.