LLM Memory & RAG 相关度: 9/10

MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling

MiniCPM Team, Wenhao An, Yingfa Chen, Yewei Fang, Jiayi Li, Xin Li, Yaohui Li, Yishan Li, Yuxuan Li, Biyuan Lin, Chuan Liu, Hezi Liu, Siyuan Liu, Hongya Lyu, Yinxu Pan, Shixin Ren, Xingyu Shen, Zhou Su, Haojun Sun, Yangang Sun, Zhen Leng Thai, Xin Tian, Rui Wang, Xiaorong Wang, Yudong Wang, Bo Wu, Xiaoyue Xu, Dong Xu, Shuaikang Xue, Jiawei Yang, Bowen Zhang, Jinqian Zhang, Letian Zhang, Shengnan Zhang, Xinyu Zhang, Xinyuan Zhang, Zhu Zhang, Hengyu Zhao, Jiacheng Zhao, Jie Zhou, Zihan Zhou, Shuo Wang, Chaojun Xiao, Xu Han, Zhiyuan Liu, Maosong Sun
arXiv: 2602.11761v1 发布: 2026-02-12 更新: 2026-02-12
下载 PDF arXiv 页面

AI 摘要

MiniCPM-SALA通过混合稀疏和线性注意力机制，在长文本建模中实现了高效的性能和内存效率。

主要贡献

  • 提出MiniCPM-SALA混合注意力架构，结合稀疏和线性注意力的优势
  • 引入成本效益高的持续训练框架，降低训练成本
  • 实现高达1M tokens的上下文长度，并提升推理速度

方法论

采用稀疏注意力(InfLLM-V2)和线性注意力(Lightning Attention)混合架构，通过层选择算法和混合位置编码(HyPE)提升效率。

原文摘要

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.

标签

长文本建模 稀疏注意力 线性注意力 模型效率 混合架构

arXiv 分类

cs.CL cs.AI cs.LG