The Warmup Dilemma: How Learning Rate Strategies Impact Speech-to-Text Model Convergence

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gaido, Marco, Papi, Sara, Bentivogli, Luisa, Brutti, Alessio, Cettolo, Mauro, Gretter, Roberto, Matassoni, Marco, Nabih, Mohamed, Negri, Matteo
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910973739139072
author Gaido, Marco
Papi, Sara
Bentivogli, Luisa
Brutti, Alessio
Cettolo, Mauro
Gretter, Roberto
Matassoni, Marco
Nabih, Mohamed
Negri, Matteo
author_facet Gaido, Marco
Papi, Sara
Bentivogli, Luisa
Brutti, Alessio
Cettolo, Mauro
Gretter, Roberto
Matassoni, Marco
Nabih, Mohamed
Negri, Matteo
contents Training large-scale models presents challenges not only in terms of resource requirements but also in terms of their convergence. For this reason, the learning rate (LR) is often decreased when the size of a model is increased. Such a simple solution is not enough in the case of speech-to-text (S2T) trainings, where evolved and more complex variants of the Transformer architecture -- e.g., Conformer or Branchformer -- are used in light of their better performance. As a workaround, OWSM designed a double linear warmup of the LR, increasing it to a very small value in the first phase before updating it to a higher value in the second phase. While this solution worked well in practice, it was not compared with alternative solutions, nor was the impact on the final performance of different LR warmup schedules studied. This paper fills this gap, revealing that i) large-scale S2T trainings demand a sub-exponential LR warmup, and ii) a higher LR in the warmup phase accelerates initial convergence, but it does not boost final performance.
format Preprint
id arxiv_https___arxiv_org_abs_2505_23420
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle The Warmup Dilemma: How Learning Rate Strategies Impact Speech-to-Text Model Convergence
Gaido, Marco
Papi, Sara
Bentivogli, Luisa
Brutti, Alessio
Cettolo, Mauro
Gretter, Roberto
Matassoni, Marco
Nabih, Mohamed
Negri, Matteo
Computation and Language
Training large-scale models presents challenges not only in terms of resource requirements but also in terms of their convergence. For this reason, the learning rate (LR) is often decreased when the size of a model is increased. Such a simple solution is not enough in the case of speech-to-text (S2T) trainings, where evolved and more complex variants of the Transformer architecture -- e.g., Conformer or Branchformer -- are used in light of their better performance. As a workaround, OWSM designed a double linear warmup of the LR, increasing it to a very small value in the first phase before updating it to a higher value in the second phase. While this solution worked well in practice, it was not compared with alternative solutions, nor was the impact on the final performance of different LR warmup schedules studied. This paper fills this gap, revealing that i) large-scale S2T trainings demand a sub-exponential LR warmup, and ii) a higher LR in the warmup phase accelerates initial convergence, but it does not boost final performance.
title The Warmup Dilemma: How Learning Rate Strategies Impact Speech-to-Text Model Convergence
topic Computation and Language
url https://arxiv.org/abs/2505.23420