Loki: Low-rank Keys for Efficient Sparse Attention

TL;DR

Loki introduces a low-rank approximation for self-attention in large language models, significantly reducing computational costs while maintaining performance by exploiting the low-dimensional structure of key vectors.

Contribution

The paper presents Loki, a novel sparse attention method that leverages low-rank key representations to improve efficiency in LLM inference.

Findings

Loki speeds up attention computation with less data movement.

Loki maintains model efficacy better than other approximation methods.

Key vectors in LLMs are low-dimensional across multiple datasets and models.

Abstract

Inference on large language models (LLMs) can be expensive in terms of the compute and memory costs involved, especially when long sequence lengths are used. In particular, the self-attention mechanism used in LLM inference contributes significantly to these costs, which has sparked an interest in approximating the self-attention computation to reduce such costs. In this work, we propose to approximate self-attention by focusing on the dimensionality of key vectors computed in the attention block. Our analysis reveals that key vectors lie in a significantly lower-dimensional space, consistently across several datasets and models. Exploiting this observation, we propose Loki, a novel sparse attention method that ranks and selects tokens in the KV-cache based on attention scores computed in low-dimensional space. Our evaluations show that Loki is able to speed up the attention computation due to reduced data movement (load/store) and compute costs while maintaining the efficacy of the models better than other popular approximation methods.