evo-model 0.2.1

DNA foundation modeling from molecular to genome scale.

Evo: DNA foundation modeling from molecular to genome scale

Evo is a biological foundation model capable of long-context modeling and design. Evo uses the StripedHyena architecture to enable modeling of sequences at a single-nucleotide, byte-level resolution with near-linear scaling of compute and memory relative to context length. Evo has 7 billion parameters and is trained on OpenGenome, a prokaryotic whole-genome dataset containing ~300 billion tokens.

We describe Evo in the paper “Sequence modeling and design from molecular to genome scale with Evo”.

We provide the following model checkpoints:

Checkpoint Name	Description
evo-1-8k-base	A model pretrained with 8,192 context. We use this model as the base model for molecular-scale finetuning tasks.
evo-1-131k-base	A model pretrained with 131,072 context using evo-1-8k-base as the base model. We use this model to reason about and generate sequences at the genome scale.
evo-1-8k-crispr	A model finetuned using evo-1-8k-base as the base model to generate CRISPR-Cas systems.
evo-1-8k-transposon	A model finetuned using evo-1-8k-base as the base model to generate IS200/IS605 transposons.

News

We identified and fixed an issue related to a wrong permutation of some projections, which affects generation quality. To use the new model revision with HuggingFace, please load as follows:

config = AutoConfig.from_pretrained(model_name, trust_remote_code=True, revision="1.1_fix")
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    config=config,
    trust_remote_code=True,
    revision="1.1_fix"
)

Contents

• Setup
  • Requirements
  • Installation
• Usage
• HuggingFace
• Together web UI
• Together API
• colab
• Playground wrapper
• Dataset
• Citation

Setup

Requirements

Evo is based on StripedHyena.

Evo uses FlashAttention-2, which may not work on all GPU architectures. Please consult the FlashAttention GitHub repository for the current list of supported GPUs.

Make sure to install the correct PyTorch version on your system.

Installation

You can install Evo using pip

pip install evo-model

or directly from the GitHub source

git clone https://github.com/evo-design/evo.git
cd evo/
pip install .

We recommend that you install the PyTorch library first, before installing all other dependencies (due to dependency issues of the flash-attn library; see, e.g., this issue).

One of our example scripts, demonstrating how to go from generating sequences with Evo to folding proteins (scripts/generation_to_folding.py), further requires the installation of prodigal. We have created an environment.yml file for this:

conda env create -f environment.yml
conda activate evo-design

Usage

Below is an example of how to download Evo and use it locally through the Python API.

from evo import Evo
import torch

device = 'cuda:0'

evo_model = Evo('evo-1-131k-base')
model, tokenizer = evo_model.model, evo_model.tokenizer
model.to(device)
model.eval()

sequence = 'ACGT'
input_ids = torch.tensor(
    tokenizer.tokenize(sequence),
    dtype=torch.int,
).to(device).unsqueeze(0)

with torch.no_grad():
    logits, _ = model(input_ids) # (batch, length, vocab)

print('Logits: ', logits)
print('Shape (batch, length, vocab): ', logits.shape)

An example of batched inference can be found in scripts/example_inference.py.

We provide an example script for how to prompt the model and sample a set of sequences given the prompt.

python -m scripts.generate \
    --model-name 'evo-1-131k-base' \
    --prompt ACGT \
    --n-samples 10 \
    --n-tokens 100 \
    --temperature 1. \
    --top-k 4 \
    --device cuda:0

We also provide an example script for using the model to score the log-likelihoods of a set of sequences.

python -m scripts.score \
    --input-fasta examples/example_seqs.fasta \
    --output-tsv scores.tsv \
    --model-name 'evo-1-131k-base' \
    --device cuda:0

HuggingFace

Evo is integrated with HuggingFace.

from transformers import AutoConfig, AutoModelForCausalLM

model_name = 'togethercomputer/evo-1-8k-base'

model_config = AutoConfig.from_pretrained(model_name, trust_remote_code=True, revision="1.1_fix")
model_config.use_cache = True

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    config=model_config,
    trust_remote_code=True,
    revision="1.1_fix"
)

Together API

Evo is available through Together AI with a web UI, where you can generate DNA sequences with a chat-like interface.

For more detailed or batch workflows, you can call the Together API with a simple example below.

import openai
import os

# Fill in your API information here.
client = openai.OpenAI(
  api_key=TOGETHER_API_KEY,
  base_url='https://api.together.xyz',
)

chat_completion = client.chat.completions.create(
  messages=[
    {
      "role": "system",
      "content": ""
    },
    {
      "role": "user",
      "content": "ACGT", # Prompt the model with a sequence.
    }
  ],
  model="togethercomputer/evo-1-131k-base",
  max_tokens=128, # Sample some number of new tokens.
  logprobs=True
)
print(
    chat_completion.choices[0].logprobs.token_logprobs,
    chat_completion.choices[0].message.content
)

Dataset

The OpenGenome dataset for pretraining Evo is available at Hugging Face datasets.

Citation

Please cite the following publication when referencing Evo.

@article{nguyen2024sequence,
   author = {Eric Nguyen and Michael Poli and Matthew G. Durrant and Brian Kang and Dhruva Katrekar and David B. Li and Liam J. Bartie and Armin W. Thomas and Samuel H. King and Garyk Brixi and Jeremy Sullivan and Madelena Y. Ng and Ashley Lewis and Aaron Lou and Stefano Ermon and Stephen A. Baccus and Tina Hernandez-Boussard and Christopher Ré and Patrick D. Hsu and Brian L. Hie },
   title = {Sequence modeling and design from molecular to genome scale with Evo},
   journal = {Science},
   volume = {386},
   number = {6723},
   pages = {eado9336},
   year = {2024},
   doi = {10.1126/science.ado9336},
   URL = {https://www.science.org/doi/abs/10.1126/science.ado9336},
}