LoGAH: Predicting 774-Million-Parameter Transformers using Graph HyperNetworks with 1/100 Parameters

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zhou, Xinyu, Knyazev, Boris, Jolicoeur-Martineau, Alexia, Fu, Jie
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866929358454652928
author Zhou, Xinyu
Knyazev, Boris
Jolicoeur-Martineau, Alexia
Fu, Jie
author_facet Zhou, Xinyu
Knyazev, Boris
Jolicoeur-Martineau, Alexia
Fu, Jie
contents A good initialization of deep learning models is essential since it can help them converge better and faster. However, pretraining large models is unaffordable for many researchers, which makes a desired prediction for initial parameters more necessary nowadays. Graph HyperNetworks (GHNs), one approach to predicting model parameters, have recently shown strong performance in initializing large vision models. Unfortunately, predicting parameters of very wide networks relies on copying small chunks of parameters multiple times and requires an extremely large number of parameters to support full prediction, which greatly hinders its adoption in practice. To address this limitation, we propose LoGAH (Low-rank GrAph Hypernetworks), a GHN with a low-rank parameter decoder that expands to significantly wider networks without requiring as excessive increase of parameters as in previous attempts. LoGAH allows us to predict the parameters of 774-million large neural networks in a memory-efficient manner. We show that vision and language models (i.e., ViT and GPT-2) initialized with LoGAH achieve better performance than those initialized randomly or using existing hypernetworks. Furthermore, we show promising transfer learning results w.r.t. training LoGAH on small datasets and using the predicted parameters to initialize for larger tasks. We provide the codes in https://github.com/Blackzxy/LoGAH .
format Preprint
id arxiv_https___arxiv_org_abs_2405_16287
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LoGAH: Predicting 774-Million-Parameter Transformers using Graph HyperNetworks with 1/100 Parameters
Zhou, Xinyu
Knyazev, Boris
Jolicoeur-Martineau, Alexia
Fu, Jie
Machine Learning
A good initialization of deep learning models is essential since it can help them converge better and faster. However, pretraining large models is unaffordable for many researchers, which makes a desired prediction for initial parameters more necessary nowadays. Graph HyperNetworks (GHNs), one approach to predicting model parameters, have recently shown strong performance in initializing large vision models. Unfortunately, predicting parameters of very wide networks relies on copying small chunks of parameters multiple times and requires an extremely large number of parameters to support full prediction, which greatly hinders its adoption in practice. To address this limitation, we propose LoGAH (Low-rank GrAph Hypernetworks), a GHN with a low-rank parameter decoder that expands to significantly wider networks without requiring as excessive increase of parameters as in previous attempts. LoGAH allows us to predict the parameters of 774-million large neural networks in a memory-efficient manner. We show that vision and language models (i.e., ViT and GPT-2) initialized with LoGAH achieve better performance than those initialized randomly or using existing hypernetworks. Furthermore, we show promising transfer learning results w.r.t. training LoGAH on small datasets and using the predicted parameters to initialize for larger tasks. We provide the codes in https://github.com/Blackzxy/LoGAH .
title LoGAH: Predicting 774-Million-Parameter Transformers using Graph HyperNetworks with 1/100 Parameters
topic Machine Learning
url https://arxiv.org/abs/2405.16287