---
language:
- en
license: mit
library_name: transformers
pipeline_tag: text-generation
tags:
- dami
- linear-attention
- long-context
- initialization
- ruler
---

# DAMI: Dual-Axis Memory Initialization for Enhanced Long-Context Retrieval in Gated Linear Attention

<div align="center">

[![License: MIT](badges/license.svg)](https://huggingface.co/SpeedAI-Lab/DAMI/blob/main/LICENSE)
[![arXiv: 2610.41994](badges/arxiv.svg)](https://arxiv.org/abs/2610.41994)
[![GitHub: Code](badges/github.svg)](https://github.com/SpeedAI-Lab/DAMI)
[![Hugging Face: Checkpoints](badges/huggingface.svg)](https://huggingface.co/SpeedAI-Lab/DAMI)

</div>

Zhangchi Feng · Harry Jiang · Yangyifei Luo · Ray Zhang · Richong Zhang

[Project page](https://speedai.com/research/dami) · [Code, evaluation data and scripts](https://cnb.cool/speed-ai/research/DAMI)

DAMI allocates memory timescales across heads and input responses across layers to improve long-context retrieval. This repository contains the 15 paper checkpoints: five architectures (GDN, KDA, GDN2, Mamba-2, RWKV-7) with three initialization conditions (Random, PoST, DAMI). Each model has approximately 100M parameters and was trained for approximately 5B tokens with a 4096-token training context.

## Checkpoints

All checkpoints are stored under `checkpoints/`, using the directory names below. Each directory contains the full model weights (`model.safetensors`), `config.json`, `generation_config.json`, `tokenizer.json`, and `tokenizer_config.json`.

| Architecture | Random | PoST | DAMI |
| --- | --- | --- | --- |
| GDN | `main_gdn_random` | `main_gdn_PoST` | `main_gdn_DAMI` |
| KDA | `main_kda_random` | `main_kda_PoST` | `main_kda_DAMI` |
| GDN2 | `main_gdn2_random` | `main_gdn2_PoST` | `main_gdn2_DAMI` |
| Mamba-2 | `main_mamba2_random` | `main_mamba2_PoST` | `main_mamba2_DAMI` |
| RWKV-7 | `main_rwkv7_random` | `main_rwkv7_PoST` | `main_rwkv7_DAMI` |

## Download and evaluation

Clone the code repository and install its dependencies as described in its README. From the code repository root, download the checkpoints:

```python
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="SpeedAI-Lab/DAMI",
    local_dir=".",
    allow_patterns=["checkpoints/**"],
)
```

To download only GDN + DAMI, use `allow_patterns=["checkpoints/main_gdn_DAMI/*"]`. Access to this repository requires authentication while it is private.

Run all 15 checkpoints on the fixed RULER evaluation data in Linux/WSL with CUDA:

```bash
python evaluate_ruler.py --preflight --all --models-root checkpoints
python evaluate_ruler.py --all --models-root checkpoints \
  --output-dir outputs/ruler-retest --skip-done
```

For a single checkpoint, replace `--all` with `--architecture gdn --initialization dami`. The evaluator includes the architecture-specific FLA and RWKV adaptations required by these checkpoints; use the provided loading path for evaluation.

The code repository includes 5,500 fixed samples across 11 RULER task/length settings and 82,500 original predictions. Evaluation uses greedy decoding with up to 128 new tokens. Scores use case-insensitive answer containment; the average weights all 11 settings equally.

## RULER results

Average accuracy (%), as reported in the paper:

| Architecture | Random | PoST | DAMI |
| --- | ---: | ---: | ---: |
| GDN | 30.42 | 33.49 | 44.36 |
| KDA | 26.38 | 33.67 | 43.09 |
| GDN2 | 30.27 | 33.47 | 43.71 |
| Mamba-2 | 9.95 | 13.73 | 24.42 |
| RWKV-7 | 23.87 | 20.69 | 29.16 |

## Citation

```bibtex
@article{feng2026dami,
  title={DAMI: Dual-Axis Memory Initialization for Enhanced Long-Context Retrieval in Gated Linear Attention},
  author={Feng, Zhangchi and Jiang, Harry and Luo, Yangyifei and Zhang, Ray and Zhang, Richong},
  journal={arXiv preprint arXiv:2610.41994},
  year={2026}
}
```

## Acknowledgement

We thank the authors and developers of [FLA](https://github.com/fla-org/flash-linear-attention), [RWKV](https://github.com/BlinkDL/RWKV-LM), [GatedDeltaNet](https://github.com/NVlabs/GatedDeltaNet), [GatedDeltaNet-2](https://github.com/NVlabs/GatedDeltaNet-2), [Kimi-Linear](https://github.com/MoonshotAI/Kimi-Linear), and [Mamba](https://github.com/state-spaces/mamba) for their contributions to the architectures and implementations used in this work. We also thank [Transformers](https://github.com/huggingface/transformers) and [FLAME](https://github.com/fla-org/flame) for their training infrastructure.

Our retrieval evaluation uses [RULER](https://github.com/NVIDIA/RULER). Benchmark attribution and data notices are provided in [data/ruler/NOTICE.md](https://github.com/SpeedAI-Lab/DAMI/blob/main/data/ruler/NOTICE.md). We thank the authors and developers for making their work available. Please cite the corresponding papers when using these resources.

## License

These checkpoints are released under the MIT License. See the LICENSE file. Benchmark attribution and data notices are provided in the code repository.