Salvato in:
Dettagli Bibliografici
Autori principali: Liu, Houjun, Bauer, John, Manning, Christopher D.
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2505.24788
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912404315570176
author Liu, Houjun
Bauer, John
Manning, Christopher D.
author_facet Liu, Houjun
Bauer, John
Manning, Christopher D.
contents Originally, dropout was seen as a breakthrough regularization technique that reduced overfitting and improved performance in almost all applications of deep learning by reducing overfitting. Yet, single-epoch pretraining tasks common to modern LLMs yield minimal overfitting, leading to dropout not being used for large LLMs. Nevertheless, no thorough empirical investigation has been done on the role of dropout in LM pretraining. Through experiments in single-epoch pretraining of both masked (BERT) and autoregressive (Pythia 160M and 1.4B) LMs with varying levels of dropout, we find that downstream performance in language modeling, morpho-syntax (BLiMP), question answering (SQuAD), and natural-language inference (MNLI) improves when dropout is not applied during pretraining. We additionally find that the recently-introduced "early dropout" also degrades performance over applying no dropout at all. We further investigate the models' editability, and find that models trained without dropout are more successful in gradient-based model editing (MEND) and equivalent in representation-based model editing (ReFT). Therefore, we advocate to drop dropout during single-epoch pretraining.
format Preprint
id arxiv_https___arxiv_org_abs_2505_24788
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Drop Dropout on Single-Epoch Language Model Pretraining
Liu, Houjun
Bauer, John
Manning, Christopher D.
Computation and Language
Artificial Intelligence
Originally, dropout was seen as a breakthrough regularization technique that reduced overfitting and improved performance in almost all applications of deep learning by reducing overfitting. Yet, single-epoch pretraining tasks common to modern LLMs yield minimal overfitting, leading to dropout not being used for large LLMs. Nevertheless, no thorough empirical investigation has been done on the role of dropout in LM pretraining. Through experiments in single-epoch pretraining of both masked (BERT) and autoregressive (Pythia 160M and 1.4B) LMs with varying levels of dropout, we find that downstream performance in language modeling, morpho-syntax (BLiMP), question answering (SQuAD), and natural-language inference (MNLI) improves when dropout is not applied during pretraining. We additionally find that the recently-introduced "early dropout" also degrades performance over applying no dropout at all. We further investigate the models' editability, and find that models trained without dropout are more successful in gradient-based model editing (MEND) and equivalent in representation-based model editing (ReFT). Therefore, we advocate to drop dropout during single-epoch pretraining.
title Drop Dropout on Single-Epoch Language Model Pretraining
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2505.24788