notesum.ai

Published at November 25

O1 Replication Journey -- Part 2: Surpassing O1-preview through Simple Distillation, Big Progress or Bitter Lesson?

cs.CL

cs.AI

Released Date: November 25, 2024

Authors: Zhen Huang¹, Haoyang Zou¹, Xuefeng Li¹, Yixiu Liu¹, Yuxiang Zheng¹, Ethan Chern¹, Shijie Xia¹, Yiwei Qin¹, Weizhe Yuan², Pengfei Liu¹

Aff.: ¹Generative AI Research Lab (GAIR); ²NYU

Arxiv: http://arxiv.org/abs/2411.16489v1

Refer to caption

Model	AIME(2024)		MATH500
Model	Accuracy	# Average Token	Accuracy	# Average Token
Proprietary
o1-preview	12/30	9083	85.5	1501
o1-mini	21/30	9903	90.0	944
Parameter Size: 72B
Ours-72B	13/30	8016	87.2	2235