Focused Large Language Models are Stable Many-Shot Learners

TL;DR

This paper introduces FocusICL, a training-free method that improves large language models' stability and performance in many-shot learning by filtering distractions and focusing attention on key content, supported by theoretical and experimental validation.

Contribution

The paper proposes FocusICL, a novel attention filtering technique for large language models that enhances many-shot learning without additional training.

Findings

FocusICL improves performance by 5.2% on average over vanilla ICL.

FocusICL scales effectively with many-shot demonstrations.

Theoretical and experimental analysis confirms attention dispersion as a key issue.

Abstract

In-Context Learning (ICL) enables large language models (LLMs) to achieve rapid task adaptation by learning from demonstrations. With the increase in available context length of LLMs, recent experiments have shown that the performance of ICL does not necessarily scale well in many-shot (demonstration) settings. We theoretically and experimentally confirm that the reason lies in more demonstrations dispersing the model attention from the query, hindering its understanding of key content. Inspired by how humans learn from examples, we propose a training-free method FocusICL, which conducts triviality filtering to avoid attention being diverted by unimportant contents at token-level and operates hierarchical attention to further ensure sufficient attention towards current query at demonstration-level. We also design an efficient hyperparameter searching strategy for FocusICL based on model perplexity of demonstrations. Comprehensive experiments validate that FocusICL achieves an average performance improvement of 5.2% over vanilla ICL and scales well with many-shot demonstrations.