Lil-Bevo

Lil-Bevo is UT Austin's submission to the BabyLM challenge, specifically the strict-small track.

Link to GitHub Repo

TLDR:

• Unigram tokenizer trained on 10M BabyLM tokens plus MAESTRO dataset for a vocab size of 16k.

• deberta-small-v3 trained on mixture of MAESTRO and 10M tokens for 5 epochs.

• Model continues training for 50 epochs on 10M tokens with sequence length of 128.

• Model is trained for 2 epochs with targeted linguistic masking with sequence length of 512.

  This README will be updated with more details soon.