evo2_7b

evo2_7b

A 7B language model for DNA that reads sequences up to 1 million letters long, from the Arc Institute's Evo 2 family. Made for researchers who work with genome data and want a mid-size checkpoint.

Made by arcinstitute on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: evo2_7b is the 7B checkpoint of Evo 2, a language model for DNA from the Arc Institute. It learns DNA the way a text model learns words, by predicting what comes next, after training on trillions of DNA tokens. This checkpoint has 32 layers and was trained on sequences up to 1 million long.

Who it suits: It suits researchers and developers who work with DNA sequences and want a model that is smaller than the 40B checkpoint but still handles extra long inputs.

Industry | Who uses it | How they use it | Example
Genomics | Researchers | Study how a model reads long DNA sequences | Feed a long sequence and inspect the output
Biotech | Data scientists | Build tools on top of a DNA language model | Wrap the model in a research pipeline
AI research | ML engineers | Compare model sizes on sequence tasks | Test the 7B checkpoint against the 1B base
Education | Students | Learn how language models treat biological data | Run a small sequence through the model

If you need a model that chats or writes prose, you can skip it, since it works on DNA sequences.

Where to use it: It runs in Python on your own computer. The card sends you to the Evo 2 GitHub project and its paper for instructions, details and examples.

• Python scripts built on the Evo 2 project code
• Research notebooks and sequence analysis pipelines
• Windows, macOS or Linux for the download step

Use case | Who | What they get back
Sequence modeling | Researchers | Model output for a DNA sequence you provide
Model comparison | Engineers | A mid-size reference point between 1B and 40B
Long-input tests | Scientists | A test of reading sequences up to 1 million long

How to set it up: You need Python and the Hugging Face hub tools. The model page gives no hardware needs or install steps, so take those from the Evo 2 GitHub project.

1. Install the Hugging Face hub tools.
2. Download the checkpoint files.
3. Get the Evo 2 code from GitHub and follow its instructions.

pip install -U huggingface_hub
hf download arcinstitute/evo2_7b

How to use it: Use the Evo 2 project's own code to load the checkpoint and pass in a DNA sequence. The page shown has no run command, so the exact call comes from that project.

Checkpoint | Layers | Parameters
evo2_40b | 50 | 40B
evo2_7b | 32 | 7B
evo2_40b_base | 50 | 40B
evo2_7b_base | 32 | 7B
evo2_1b_base | 25 | 1B

Example: You load evo2_7b with the Evo 2 project's code and give it a DNA sequence. The model returns its predictions for the sequence. The card shows no sample output, so results come from your own run.

For newcomers and experts: A DNA language model treats a genome like text and learns the patterns in it. Begin with the project's examples before using your own data.

The two large checkpoints, evo2_40b and evo2_7b, were trained up to 1 million sequence length. The base checkpoints in 40B, 7B and 1B sizes were trained on a shorter 8192 context length.

Get a DNA sequence
Load the checkpoint with the Evo 2 code
The model reads the sequence
You inspect its predictions

Good to know: Evo 2 comes in several sizes, and this page is the 7B one with the long context.

Title: Layers in each Evo 2 size
40B checkpoint: 50 layers
7B checkpoint: 32 layers
1B base checkpoint: 25 layers

• Context: this checkpoint was trained up to 1 million sequence length, while the base ones used 8192.
• Versions: the page lists 1 quantized and 4 fine-tuned versions made by other people.
• License: Apache 2.0.
• Cost to run: none named.
• Data it reaches: the DNA sequences you give it, handled on your own machine.
• Not for: text chat or writing, since the model reads DNA sequences.

On Hugging Face, 13 November 2026

• Downloads in the last 30 days: 1,601
• Likes: 87
• License: apache-2.0
• Quantized versions: 1
• Fine-tunes: 4

Links

• Model: https://huggingface.co/arcinstitute/evo2_7b
• Files: https://huggingface.co/arcinstitute/evo2_7b/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download arcinstitute/evo2_7b

Then follow the usage in its model card. Nothing in this kit runs by itself.