Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915761386160128 |
|---|---|
| author | Chen, Yingfa Thai, Zhen Leng Zhou, Zihan Zhang, Zhu Shen, Xingyu Wang, Shuo Xiao, Chaojun Han, Xu Liu, Zhiyuan |
| author_facet | Chen, Yingfa Thai, Zhen Leng Zhou, Zihan Zhang, Zhu Shen, Xingyu Wang, Shuo Xiao, Chaojun Han, Xu Liu, Zhiyuan |
| contents | Hybrid Transformer architectures, which combine softmax attention blocks and recurrent neural networks (RNNs), have shown a desirable performance-throughput tradeoff for long-context modeling, but their adoption and studies are hindered by the prohibitive cost of large-scale pre-training from scratch. Some recent studies have shown that pre-trained softmax attention blocks can be converted into RNN blocks through parameter transfer and knowledge distillation. However, these transfer methods require substantial amounts of training data (more than 10B tokens), and the resulting hybrid models also exhibit poor long-context performance, which is the scenario where hybrid models enjoy significant inference speedups over Transformer-based models. In this paper, we present HALO (Hybrid Attention via Layer Optimization), a pipeline for distilling Transformer models into RNN-attention hybrid models. We then present HypeNet, a hybrid architecture with superior length generalization enabled by a novel position encoding scheme (named HyPE) and various architectural modifications. We convert the Qwen3 series into HypeNet using HALO, achieving performance comparable to the original Transformer models while enjoying superior long-context performance and efficiency. The conversion requires just 2.3B tokens, less than 0.01% of their pre-training data |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_22156 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts Chen, Yingfa Thai, Zhen Leng Zhou, Zihan Zhang, Zhu Shen, Xingyu Wang, Shuo Xiao, Chaojun Han, Xu Liu, Zhiyuan Computation and Language Artificial Intelligence Machine Learning Hybrid Transformer architectures, which combine softmax attention blocks and recurrent neural networks (RNNs), have shown a desirable performance-throughput tradeoff for long-context modeling, but their adoption and studies are hindered by the prohibitive cost of large-scale pre-training from scratch. Some recent studies have shown that pre-trained softmax attention blocks can be converted into RNN blocks through parameter transfer and knowledge distillation. However, these transfer methods require substantial amounts of training data (more than 10B tokens), and the resulting hybrid models also exhibit poor long-context performance, which is the scenario where hybrid models enjoy significant inference speedups over Transformer-based models. In this paper, we present HALO (Hybrid Attention via Layer Optimization), a pipeline for distilling Transformer models into RNN-attention hybrid models. We then present HypeNet, a hybrid architecture with superior length generalization enabled by a novel position encoding scheme (named HyPE) and various architectural modifications. We convert the Qwen3 series into HypeNet using HALO, achieving performance comparable to the original Transformer models while enjoying superior long-context performance and efficiency. The conversion requires just 2.3B tokens, less than 0.01% of their pre-training data |
| title | Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2601.22156 |