notesum.ai

Published at October 29

SVIP: Towards Verifiable Inference of Open-source Large Language Models

cs.DS

cs.AI

cs.CR

cs.LG

stat.ML

Released Date: October 29, 2024

Authors: Yifan Sun¹, Yuhang Li¹, Yue Zhang², Yuchen Jin², Huan Zhang¹

Aff.: ¹University of Illinois Urbana-Champaign; ²Hyperbolic Labs

Arxiv: https://arxiv.org/abs/2410.22307v1

Refer to caption

Specified Model	FNR $\downarrow$	FPR $\downarrow$
Specified Model	FNR $\downarrow$	Random	GPT2-XL	GPT-NEO-2.7B	GPT-J-6B	OPT-6.7B	Vicuna-7B	Llama-2-7B
Llama-2-13B	4.41%	1.97%	1.90%	1.77%	1.75%	2.03%	2.44%	2.04%
GPT-NeoX-20B	3.47%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%	0.00%
OPT-30B	3.42%	0.05%	0.33%	0.61%	0.47%	0.83%	0.34%	0.35%
Falcon-40B	3.02%	0.00%	0.00%	0.01%	0.00%	0.00%	0.00%	0.00%
Llama-3.1-70B	3.13%	0.26%	1.97%	1.04%	1.98%	2.07%	0.90%	0.81%