Saved in:
Bibliographic Details
Main Authors: Song, Yuda, Zhang, Hanlin, Eisenach, Carson, Kakade, Sham, Foster, Dean, Ghai, Udaya
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.02674
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917935787802624
author Song, Yuda
Zhang, Hanlin
Eisenach, Carson
Kakade, Sham
Foster, Dean
Ghai, Udaya
author_facet Song, Yuda
Zhang, Hanlin
Eisenach, Carson
Kakade, Sham
Foster, Dean
Ghai, Udaya
contents Self-improvement is a mechanism in Large Language Model (LLM) pre-training, post-training and test-time inference. We explore a framework where the model verifies its own outputs, filters or reweights data based on this verification, and distills the filtered data. Despite several empirical successes, a fundamental understanding is still lacking. In this work, we initiate a comprehensive, modular and controlled study on LLM self-improvement. We provide a mathematical formulation for self-improvement, which is largely governed by a quantity which we formalize as the generation-verification gap. Through experiments with various model families and tasks, we discover a scaling phenomenon of self-improvement -- a variant of the generation-verification gap scales monotonically with the model pre-training flops. We also examine when self-improvement is possible, an iterative self-improvement procedure, and ways to improve its performance. Our findings not only advance understanding of LLM self-improvement with practical implications, but also open numerous avenues for future research into its capabilities and boundaries.
format Preprint
id arxiv_https___arxiv_org_abs_2412_02674
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models
Song, Yuda
Zhang, Hanlin
Eisenach, Carson
Kakade, Sham
Foster, Dean
Ghai, Udaya
Computation and Language
Machine Learning
Self-improvement is a mechanism in Large Language Model (LLM) pre-training, post-training and test-time inference. We explore a framework where the model verifies its own outputs, filters or reweights data based on this verification, and distills the filtered data. Despite several empirical successes, a fundamental understanding is still lacking. In this work, we initiate a comprehensive, modular and controlled study on LLM self-improvement. We provide a mathematical formulation for self-improvement, which is largely governed by a quantity which we formalize as the generation-verification gap. Through experiments with various model families and tasks, we discover a scaling phenomenon of self-improvement -- a variant of the generation-verification gap scales monotonically with the model pre-training flops. We also examine when self-improvement is possible, an iterative self-improvement procedure, and ways to improve its performance. Our findings not only advance understanding of LLM self-improvement with practical implications, but also open numerous avenues for future research into its capabilities and boundaries.
title Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2412.02674