notesum.ai

Published at November 26

One Mind, Many Tongues: A Deep Dive into Language-Agnostic Knowledge Neurons in Large Language Models

cs.CL

Released Date: November 26, 2024

Authors: Pengfei Cao, Yuheng Chen, Zhuoran Jin, Yubo Chen, Kang Liu, Jun Zhao

Arxiv: http://arxiv.org/abs/2411.17401v1

Refer to caption

Base Models	Methods	Enhance	Suppress	Average
mGPT	MaTrice	1.000	1.000	1.000
	w/o UaQ	0.830	0.850	0.840
	w/o UaL	0.910	0.920	0.915
LLaMA2	MaTrice	1.000	1.000	1.000
	w/o UaQ	0.870	0.860	0.865
	w/o UaL	0.930	0.940	0.935
LLaMA3	MaTrice	1.000	1.000	1.000
	w/o UaQ	0.850	0.840	0.845
	w/o UaL	0.920	0.930	0.925
mBERT	MaTrice	1.000	1.000	1.000
	w/o UaQ	0.820	0.830	0.825
	w/o UaL	0.900	0.910	0.905