Non-Vacuous Generalization Bounds for Large Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lotfi, Sanae, Finzi, Marc, Kuang, Yilun, Rudner, Tim G. J., Goldblum, Micah, Wilson, Andrew Gordon
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914873766576128
author Lotfi, Sanae
Finzi, Marc
Kuang, Yilun
Rudner, Tim G. J.
Goldblum, Micah
Wilson, Andrew Gordon
author_facet Lotfi, Sanae
Finzi, Marc
Kuang, Yilun
Rudner, Tim G. J.
Goldblum, Micah
Wilson, Andrew Gordon
contents Modern language models can contain billions of parameters, raising the question of whether they can generalize beyond the training data or simply parrot their training corpora. We provide the first non-vacuous generalization bounds for pretrained large language models (LLMs), indicating that language models are capable of discovering regularities that generalize to unseen data. In particular, we derive a compression bound that is valid for the unbounded log-likelihood loss using prediction smoothing, and we extend the bound to handle subsampling, accelerating bound computation by orders of magnitude on massive datasets. To achieve the extreme level of compression required for non-vacuous bounds, we devise SubLoRA, a simple low-dimensional nonlinear parameterization that leads to non-vacuous generalization bounds for models with nearly a billion parameters. Finally, we use our bounds to understand LLM generalization and find that larger models have better generalization bounds and are more compressible than smaller models.
format Preprint
id arxiv_https___arxiv_org_abs_2312_17173
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Non-Vacuous Generalization Bounds for Large Language Models
Lotfi, Sanae
Finzi, Marc
Kuang, Yilun
Rudner, Tim G. J.
Goldblum, Micah
Wilson, Andrew Gordon
Machine Learning
Modern language models can contain billions of parameters, raising the question of whether they can generalize beyond the training data or simply parrot their training corpora. We provide the first non-vacuous generalization bounds for pretrained large language models (LLMs), indicating that language models are capable of discovering regularities that generalize to unseen data. In particular, we derive a compression bound that is valid for the unbounded log-likelihood loss using prediction smoothing, and we extend the bound to handle subsampling, accelerating bound computation by orders of magnitude on massive datasets. To achieve the extreme level of compression required for non-vacuous bounds, we devise SubLoRA, a simple low-dimensional nonlinear parameterization that leads to non-vacuous generalization bounds for models with nearly a billion parameters. Finally, we use our bounds to understand LLM generalization and find that larger models have better generalization bounds and are more compressible than smaller models.
title Non-Vacuous Generalization Bounds for Large Language Models
topic Machine Learning
url https://arxiv.org/abs/2312.17173