Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chang, Ting-Yun, Thomason, Jesse, Jia, Robin
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2406.13131
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916424271790080
author Chang, Ting-Yun
Thomason, Jesse
Jia, Robin
author_facet Chang, Ting-Yun
Thomason, Jesse
Jia, Robin
contents This paper studies in-context learning by decomposing the output of large language models into the individual contributions of attention heads and MLPs (components). We observe curious components: good-performing ones that individually do well on a classification task, even when the model performs poorly; bad-performing ones that do much worse than chance; and label-biased components that always predict the same label. We find that component accuracies are well-correlated across different demonstration sets and perturbations of prompt templates. Based on our findings, we propose component reweighting, which learns to linearly re-scale the component activations from a few labeled examples. Given 24 labeled examples, our method improves by an average of 6.0% accuracy points over 24-shot ICL across 8 tasks on Llama-2-7B. Overall, this paper both enriches our understanding of ICL and provides a practical method for improvement by examining model internals.
format Preprint
id arxiv_https___arxiv_org_abs_2406_13131
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Models
Chang, Ting-Yun
Thomason, Jesse
Jia, Robin
Computation and Language
This paper studies in-context learning by decomposing the output of large language models into the individual contributions of attention heads and MLPs (components). We observe curious components: good-performing ones that individually do well on a classification task, even when the model performs poorly; bad-performing ones that do much worse than chance; and label-biased components that always predict the same label. We find that component accuracies are well-correlated across different demonstration sets and perturbations of prompt templates. Based on our findings, we propose component reweighting, which learns to linearly re-scale the component activations from a few labeled examples. Given 24 labeled examples, our method improves by an average of 6.0% accuracy points over 24-shot ICL across 8 tasks on Llama-2-7B. Overall, this paper both enriches our understanding of ICL and provides a practical method for improvement by examining model internals.
title When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Models
topic Computation and Language
url https://arxiv.org/abs/2406.13131