Romil Shah
Results
2
issues of
Romil Shah
- Changes made to `train.py`, `model.py`, and `norms.py` to check if Transformer Engine can be imported for P5 instances or H100s and use FP8 for Linear and LayerNorm layers. -...
- Adding flags for seamless S3 streaming using BytesIO - Adding LLamaTokenizer for using LLama based tokenizers