README.md

---
license: apache-2.0
tags:
  - biology
  - genomics
  - dna
  - mamba
  - masked-lm
library_name: pytorch
---

# LDARNet-110M

Pretrained LDARNet (~110M params) with learnable DNA tokenization (dynamic chunking + BiMamba-2).

- **Paper:** [arXiv:2606.04552](https://arxiv.org/abs/2606.04552)
- **Code:** [ICML-LDARNet](https://github.com/darlednik/ICML-LDARNet)

## Files

- `model_ckpt_110m.pt` — MLM checkpoint with embedded `LDarConfig`

## Download

Clone the code repo and install dependencies, then download the weights:

```bash
huggingface-cli download darlednik/LDARNet-110M model_ckpt_110m.pt --local-dir models_ckpts
```

## Load

```python
import torch
from ldar.utils.ckpt import load_ldar_from_ckpt

model, cfg = load_ldar_from_ckpt(
    "models_ckpts/model_ckpt_110m.pt",
    device="cuda",
    dtype=torch.bfloat16,
)
```

## Architecture

| Component | Layout | `d_model` |
|---|---|---|
| Encoder | `m3t1` — 3× BiMamba-2 + 1 local-attention layer | 512 |
| Backbone | `M10` — 10× BiMamba-2 (+ SwiGLU) | 768 |
| Decoder | `m4` — 4× BiMamba-2 | 512 |

- Compression ratio **N = 4**
- Byte vocabulary: `{A, C, G, T, N, [MASK], <pad>}`

## Citation

```bibtex
@misc{ledneva2026ldarnetdnaadaptiverepresentation,
      title={LDARNet: DNA Adaptive Representation Network with Learnable Tokenization for Genomic Modeling},
      author={Daria Ledneva and Denis Kuznetsov},
      year={2026},
      eprint={2606.04552},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.04552},
}
```