Fortify the Shortest Stave in Attention: Enhancing Context Awareness of Large Language Models for Effective Tool Use
This addresses the issue of context awareness in LLMs for tasks like tool-use and RAG, offering a significant performance boost, though it is an incremental improvement on existing attention mechanisms.
The paper tackles the problem of large language models overlooking crucial contextual information due to inherent attention waveform patterns, particularly in tool-use tasks, by proposing Attention Buckets, a novel inference method that processes input through multiple parallel processes with distinct rotary position embedding angles to compensate attention troughs; this elevates a 7B model to achieve state-of-the-art performance comparable to GPT-4 on the largest tool-use benchmark.
In this paper, we demonstrate that an inherent waveform pattern in the attention allocation of large language models (LLMs) significantly affects their performance in tasks demanding a high degree of context awareness, such as utilizing LLMs for tool-use. Specifically, the crucial information in the context will be potentially overlooked by model when it is positioned in the trough zone of the attention waveform, leading to decreased performance. To address this issue, we propose a novel inference method named Attention Buckets. It allows LLMs to process their input through multiple parallel processes. Each process utilizes a distinct base angle for the rotary position embedding, thereby creating a unique attention waveform. By compensating an attention trough of a particular process with an attention peak of another process, our approach enhances LLM's awareness to various contextual positions, thus mitigating the risk of overlooking crucial information. In the largest tool-use benchmark, our method elevates a 7B model to achieve state-of-the-art performance, comparable to that of GPT-4. On other benchmarks and some RAG tasks, which also demand a thorough understanding of contextual content, Attention Buckets also exhibited notable enhancements in performance.