MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling
Abstract
The evolution of large language models (LLMs) towards applications with ultra-long contexts faces challenges posed by the high computational and memory costs of the Transformer architecture. While existing sparse and linear attention mechanisms attempt to mitigate these issues, they typically involve a trade-off between memory efficiency and model performance. This paper introduces MiniCPM-SALA, a 9B-parameter hybrid architecture that integrates the high-fidelity long-context modeling of sparse attention (InfLLM-V2) with the global efficiency of linear attention (Lightning Attention). By employing a layer selection algorithm to integrate these mechanisms in a 1:3 ratio and utilizing a hybrid positional encoding (HyPE), the model maintains efficiency and performance for long-context tasks. Furthermore, we introduce a cost-effective continual training framework that transforms pre-trained Transformer-based models into hybrid models, which reduces training costs by approximately 75% compared to training from scratch. Extensive experiments show that MiniCPM-SALA maintains general capabilities comparable to full-attention models while offering improved efficiency. On a single NVIDIA A6000D GPU, the model achieves up to 3.5x the inference speed of the full-attention model at the sequence length of 256K tokens and supports context lengths of up to 1M tokens, a scale where traditional full-attention 8B models fail because of memory constraints.
Cite
@article{arxiv.2602.11761,
title = {MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling},
author = {MiniCPM Team and Wenhao An and Yingfa Chen and Yewei Fang and Jiayi Li and Xin Li and Yaohui Li and Yishan Li and Yuxuan Li and Biyuan Lin and Chuan Liu and Hezi Liu and Siyuan Liu and Hongya Lyu and Yinxu Pan and Shixin Ren and Xingyu Shen and Zhou Su and Haojun Sun and Yangang Sun and Zhen Leng Thai and Xin Tian and Rui Wang and Xiaorong Wang and Yudong Wang and Bo Wu and Xiaoyue Xu and Dong Xu and Shuaikang Xue and Jiawei Yang and Bowen Zhang and Jinqian Zhang and Letian Zhang and Shengnan Zhang and Xinyu Zhang and Xinyuan Zhang and Zhu Zhang and Hengyu Zhao and Jiacheng Zhao and Zhi Zheng and Jie Zhou and Zihan Zhou and Shuo Wang and Chaojun Xiao and Xu Han and Zhiyuan Liu and Maosong Sun},
journal= {arXiv preprint arXiv:2602.11761},
year = {2026}
}
Comments
MiniCPM-SALA Technical Report