> ## Documentation Index
> Fetch the complete documentation index at: https://proto.evodesign.org/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Evo2 DNA Language Model
> Evo2 genome language model for DNA sequence generation
**License:** Evo2 is open source and free for academic and commercial use under an Apache-2.0 license. Please refer to [the license](https://github.com/arcinstitute/evo2/blob/main/LICENSE) for full terms.
This generator is open source. Any third-party models, product names, or trademarks referenced are the property of their respective owners, and Proto is not affiliated with them.
```bibtex theme={null}
@ARTICLE{Brixi2026-jn,
title = "Genome modelling and design across all domains of life with Evo 2",
author = "Brixi, Garyk and Durrant, Matthew G and Ku, Jerome and
Naghipourfar, Mohsen and Poli, Michael and Sun, Gwanggyu and
Brockman, Greg and Chang, Daniel and Fanton, Alison and Gonzalez,
Gabriel A and King, Samuel H and Li, David B and Merchant, Aditi
T and Nguyen, Eric and Ricci-Tam, Chiara and Romero, David W and
Schmok, Jonathan C and Taghibakhshi, Ali and Vorontsov, Anton and
Yang, Brandon and Deng, Myra and Gorton, Liv and Nguyen, Nam and
Wang, Nicholas K and Pearce, Michael T and Simon, Elana and
Adams, Etowah and Amador, Zachary J and Ashley, Euan A and
Baccus, Stephen A and Dai, Haoyu and Dillmann, Steven and Ermon,
Stefano and Guo, Daniel and Herschl, Michael H and Ilango, Rajesh
and Janik, Ken and Lu, Amy X and Mehta, Reshma and Mofrad,
Mohammad R K and Ng, Madelena Y and Pannu, Jaspreet and R{\'e},
Christopher and St John, John and Sullivan, Jeremy and Tey,
Joseph and Viggiano, Ben and Zhu, Kevin and Zynda, Greg and
Balsam, Daniel and Collison, Patrick and Costa, Anthony B and
Hernandez-Boussard, Tina and Ho, Eric and Liu, Ming-Yu and
McGrath, Thomas and Powell, Kimberly and Pinglay, Sudarshan and
Burke, Dave P and Goodarzi, Hani and Hsu, Patrick D and Hie,
Brian L",
journal = "Nature",
publisher = "Springer Science and Business Media LLC",
pages = "1--13",
doi = "10.1038/s41586-026-10176-5",
month = mar,
year = 2026,
language = "en"
}
```
Sequence generator using Evo2 genomic language model for DNA generation.
This generator uses the Evo2 7B parameter model to autoregressively generate
DNA sequences from prompt sequences. Supports advanced sampling strategies,
KV caching for efficiency, and batch generation.
The generator category is `"autoregressive"`, indicating sequences
are generated token-by-token from left to right.
The number of tokens to generate is automatically calculated based on the
assigned segment's sequence\_length, prompt length, and prepend\_prompt setting.
## API Reference
Configuration object for Evo2Generator.
This class defines configuration parameters for the Evo2 generator, which uses
a 7B parameter genomic language model to generate DNA sequences autoregressively
from prompt sequences.
All prompts must have identical lengths for batched generation. For detailed
information on Evo2 parameters, see: [https://github.com/arcinstitute/evo2](https://github.com/arcinstitute/evo2)
Prompt sequences for DNA sequence generation (single prompt or multiple)
Evo2 model variant to load (currently only evo2\_7b).
Options: `evo2_7b`, `evo2_20b`, `evo2_40b`, `evo2_7b_base`, `evo2_40b_base`, `evo2_1b_base`, `evo2_7b_262k`, `evo2_7b_microviridae`
Path to local checkpoint weights for custom or finetuned models
GPU device to run Evo2 on (e.g. 'cuda' or 'cuda:0').
Limits sampling to the top-k most probable tokens at each generation step.
Nucleus sampling cutoff. Restricts to the smallest token set with cumulative prob ≥ top-p.
Sharpness of sampling. Below 1 favors high-probability tokens; above 1 increases diversity.
Optional number of tokens to prefill in parallel before switching to prompt forcing.
Optional maximum sequence length to generate. Determines the max size of the cache if larger.
Whether to stop at end-of-sequence token
Generate all prompts together in a single batched forward pass. Required for multiple prompts.
Number of sequences to process simultaneously on GPU
Whether to reuse KV-cache state across decoding steps to avoid recomputation.
Retain and expose the per-sequence KV-cache after generation so downstream callers can continue.
Whether to prepend prompt to generation
Whether to print verbose output
## Usage
```python python icon="python" theme={null}
>>> from proto_language.generator import Evo2Generator, Evo2GeneratorConfig
>>> from proto_language.core import Segment, SequenceType
>>> config = Evo2GeneratorConfig(prompts="ATG", temperature=0.8)
>>> gen = Evo2Generator(config)
>>> # Segment length determines how many tokens to generate
>>> segment = Segment(length=1003, sequence_type="dna")
>>> gen.assign(segment) # prepend_prompt defaults to False, so max_new_tokens = 1003
>>> gen.sample() # Generates DNA sequences
```
## Metadata
| Property | Value |
| ------------------------ | ---------------- |
| Key | `evo2` |
| Class | `Evo2Generator` |
| Category | `autoregressive` |
| Input Type | `prompt` |
| Uses GPU | `True` |
| Supported Sequence Types | `dna` |
| Allows Empty Start | `False` |
evo-design/proto-language/proto\_language/generator/evo2\_generator.py
dguo8412