RITA: a Study on Scaling Up Generative Protein Sequence Models

RITA is a family of autoregressive protein models, developed by a collaboration of Lighton, the OATML group at Oxford, and the Debbie Marks Lab at Harvard.

Model	#Params	d_model	layers	lm loss uniref-100
Small	85M	768	12	2.31
Medium	300M	1024	24	2.01
Large	680M	1536	24	1.82
XLarge	1.2B	2048	24	1.70

Results

For full results see our preprint: https://arxiv.org/abs/2205.05789

Usage

Instantiate a model like so:

from transformers import AutoModel, AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("lightonai/RITA_s", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("lightonai/RITA_s")

for generation we support pipelines:

from transformers import pipeline
rita_gen = pipeline('text-generation', model=model, tokenizer=tokenizer)
sequences = rita_gen("MAB", max_length=20, do_sample=True, top_k=950, repetition_penalty=1.2, 
                     num_return_sequences=2, eos_token_id=2)
for seq in sequences:
    print(f"seq: {seq['generated_text'].replace(' ', '')}")

Or see example.py

How to cite

@article{hesslow2022rita,
  title={RITA: a Study on Scaling Up Generative Protein Sequence Models},
  author={Hesslow, Daniel and Zanichelli, Niccol{\'o} and Notin, Pascal and Poli, Iacopo and Marks, Debora},
  journal={arXiv preprint arXiv:2205.05789},
  year={2022}
}