dbert2_170 / README.md
vinayh19's picture
Update README.md
16a3bfa verified
metadata
license: apache-2.0
language:
  - en
tags:
  - biology
  - medical
  - code
  - NLP
  - Genomics

DNABERT-2 MosaicBERT Architecture; created new model from scratch

Converted from Composer checkpoint.

This model build uses Flash Attention 2 and ignores triton; the max_seq_len parameter is set to 170 and trained using amp_bf16 precision parameter.