Understanding The Rope Extensions Of Long-context Llms: An Attention Perspective

Zhong Meizhi, Zhang Chen, Lei Yikun, Liu Xikai, Gao Yan, Hu Yao, Chen Kehai, Zhang Min. Arxiv 2024

[Paper]    
Attention Mechanism Model Architecture Pretraining Methods Training Techniques

Enabling LLMs to handle lengthy context is currently a research hotspot. Most LLMs are built upon rotary position embedding (RoPE), a popular position encoding method. Therefore, a prominent path is to extrapolate the RoPE trained on comparably short texts to far longer texts. A heavy bunch of efforts have been dedicated to boosting the extrapolation via extending the formulations of the RoPE, however, few of them have attempted to showcase their inner workings comprehensively. In this paper, we are driven to offer a straightforward yet in-depth understanding of RoPE extensions from an attention perspective and on two benchmarking tasks. A broad array of experiments reveals several valuable findings: 1) Maintaining attention patterns to those at the pretrained length improves extrapolation; 2) Large attention uncertainty leads to retrieval errors; 3) Using longer continual pretraining lengths for RoPE extensions could reduce attention uncertainty and significantly enhance extrapolation.

Similar Work