---
tags:
- dna
- human_genome
---

# GENA-LM (gena-lm-bigbird-base-sparse-t2t)

GENA-LM is a Family of Open-Source Foundational Models for Long DNA Sequences.

GENA-LM models are transformer masked language models trained on human DNA sequence. 

`gena-lm-bigbird-base-sparse-t2t` follows the BigBird architecture and uses sparse attention from DeepSpeed.

Differences between GENA-LM (`gena-lm-bigbird-base-sparse-t2t`) and DNABERT:
- BPE tokenization instead of k-mers;
- input sequence size is about 36000 nucleotides (4096 BPE tokens) compared to 512 nucleotides of DNABERT;
- pre-training on T2T vs. GRCh38.p13 human genome assembly.

Source code and data: https://github.com/AIRI-Institute/GENA_LM

Paper: https://academic.oup.com/nar/article/53/2/gkae1310/7954523

## Installation
`gena-lm-bigbird-base-sparse-t2t` sparse ops require DeepSpeed.

### DeepSpeed
DeepSpeed installation is needed to work with SparseAttention versions of language models. DeepSpeed Sparse attention supports only GPUs with compute compatibility >= 7 (V100, T4, A100).
```bash
pip install triton==1.0.0
DS_BUILD_SPARSE_ATTN=1 pip install deepspeed==0.6.0 --global-option="build_ext" --global-option="-j8" --no-cache
```
and check installation with
```bash
ds_report
```

### APEX for FP16
Install APEX https://github.com/NVIDIA/apex#quick-start
```
git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
```

## Examples

### How to load pre-trained model for Masked Language Modeling
```python
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t')
model = AutoModel.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t', trust_remote_code=True)

```

### How to load pre-trained model to fine-tune it on classification task
Get model class from GENA-LM repository:
```bash
git clone https://github.com/AIRI-Institute/GENA_LM.git
```

```python
from GENA_LM.src.gena_lm.modeling_bert import BertForSequenceClassification
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t')
model = BertForSequenceClassification.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t')
```
or you can just download [modeling_bert.py](https://github.com/AIRI-Institute/GENA_LM/tree/main/src/gena_lm) and put it close to your code.

OR you can get model class from HuggingFace AutoModel:
```python
from transformers import AutoTokenizer, AutoModel
model = AutoModel.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t', trust_remote_code=True)
gena_module_name = model.__class__.__module__
print(gena_module_name)
import importlib
# available class names:
# - BertModel, BertForPreTraining, BertForMaskedLM, BertForNextSentencePrediction,
# - BertForSequenceClassification, BertForMultipleChoice, BertForTokenClassification,
# - BertForQuestionAnswering
# check https://huggingface.co/docs/transformers/model_doc/bert
cls = getattr(importlib.import_module(gena_module_name), 'BertForSequenceClassification')
print(cls)
model = cls.from_pretrained('AIRI-Institute/gena-lm-bigbird-base-sparse-t2t', num_labels=2)
```

## Model description
GENA-LM (`gena-lm-bigbird-base-sparse-t2t`) model is trained in a masked language model (MLM) fashion, following the methods proposed in the BigBird paper by masking 15% of tokens. Model config for `gena-lm-bigbird-base-sparse-t2t` is similar to the `google/bigbird-roberta-base`:

- 4096 Maximum sequence length
- 12 Layers, 12 Attention heads
- 768 Hidden size
- sparse config:
    - block size: 64
    - random blocks: 3
    - global blocks: 2
    - sliding window blocks: 3
- Rotary positional embeddings
- 32k Vocabulary size, tokenizer trained on DNA data.

We pre-trained `gena-lm-bigbird-base-sparse-t2t` using the latest T2T human genome assembly (https://www.ncbi.nlm.nih.gov/assembly/GCA_009914755.3/). The data was augmented by sampling mutations from 1000-genome SNPs (gnomAD dataset). Pre-training was performed for 800,000 iterations with batch size 256. We modified Transformer with [Pre-Layer normalization](https://arxiv.org/abs/2002.04745).

## Evaluation
For evaluation results, see our paper: https://academic.oup.com/nar/article/53/2/gkae1310/7954523

## Citation
```bibtex
@article{GENA_LM,
    author = {Fishman, Veniamin and Kuratov, Yuri and Shmelev, Aleksei and Petrov, Maxim and Penzar, Dmitry and Shepelin, Denis and Chekanov, Nikolay and Kardymon, Olga and Burtsev, Mikhail},
    title = {GENA-LM: a family of open-source foundational DNA language models for long sequences},
    journal = {Nucleic Acids Research},
    volume = {53},
    number = {2},
    pages = {gkae1310},
    year = {2025},
    month = {01},
    issn = {0305-1048},
    doi = {10.1093/nar/gkae1310},
    url = {https://doi.org/10.1093/nar/gkae1310},
    eprint = {https://academic.oup.com/nar/article-pdf/53/2/gkae1310/61443229/gkae1310.pdf},
}
```