evo2

1# Evo 2 — DNA Language Model
2 
3## Prerequisites
4 
5| Requirement | Minimum         | Recommended  |
6| ----------- | --------------- | ------------ |
7| Python      | 3.11            | 3.12 (<3.13) |
8| CUDA        | 12.1+           | 12.4+        |
9| GPU VRAM    | 24 GB (7B bf16) | 80 GB (40B)  |
10| RAM         | 32 GB           | 128 GB       |
11 
12## How to run
13 
14### Installation
15 
16```bash
17pip install evo2
18# Weights pulled from Hugging Face on first model load.
19```
20 
21### Loading and scoring
22 
23```python
24from evo2 import Evo2
25 
26model = Evo2("evo2_7b")        # or "evo2_40b" — see model table
27seqs = ["ATCG" * 50, "GGGCTTAA" * 25]
28ll = model.score_sequences(seqs)   # → list[float], mean per-token log-likelihood
29print(ll)
30```
31 
32### Generation
33 
34```python
35out = model.generate(
36    prompt_seqs=["ATGAAAGCT"],
37    n_tokens=256,
38    temperature=0.7,
39)
40print(out.sequences[0])
41```
42 
43## Models
44 
45| Name           | Params | Context | VRAM (bf16) | Notes                                |
46| -------------- | ------ | ------- | ----------- | ------------------------------------ |
47| `evo2_7b`      | 7 B    | 1 M nt  | ~22 GB      | Default; fits on a single 24 GB+ GPU |
48| `evo2_40b`     | 40 B   | 1 M nt  | ~78 GB      | H100 80 GB or multi-GPU              |
49| `evo2_1b_base` | 1 B    | 8 K nt  | ~6 GB       | FP8 path requires sm_89+ (H100)      |
50 
51## Output format
52 
53`score_sequences` returns a `list[float]` (or `np.ndarray`) of mean log-likelihoods,
54…