Dataless Knowledge Fusion by Merging Weights of Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Jin, Xisen, Ren, Xiang, Preotiuc-Pietro, Daniel, Cheng, Pengxiang
Formato: Preprint
Publicado: 2022
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910960089825280
author Jin, Xisen
Ren, Xiang
Preotiuc-Pietro, Daniel
Cheng, Pengxiang
author_facet Jin, Xisen
Ren, Xiang
Preotiuc-Pietro, Daniel
Cheng, Pengxiang
contents Fine-tuning pre-trained language models has become the prevalent paradigm for building downstream NLP models. Oftentimes fine-tuned models are readily available but their training data is not, due to data privacy or intellectual property concerns. This creates a barrier to fusing knowledge across individual models to yield a better single model. In this paper, we study the problem of merging individual models built on different training data sets to obtain a single model that performs well both across all data set domains and can generalize on out-of-domain data. We propose a dataless knowledge fusion method that merges models in their parameter space, guided by weights that minimize prediction differences between the merged model and the individual models. Over a battery of evaluation settings, we show that the proposed method significantly outperforms baselines such as Fisher-weighted averaging or model ensembling. Further, we find that our method is a promising alternative to multi-task learning that can preserve or sometimes improve over the individual models without access to the training data. Finally, model merging is more efficient than training a multi-task model, thus making it applicable to a wider set of scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2212_09849
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Dataless Knowledge Fusion by Merging Weights of Language Models
Jin, Xisen
Ren, Xiang
Preotiuc-Pietro, Daniel
Cheng, Pengxiang
Computation and Language
Machine Learning
Fine-tuning pre-trained language models has become the prevalent paradigm for building downstream NLP models. Oftentimes fine-tuned models are readily available but their training data is not, due to data privacy or intellectual property concerns. This creates a barrier to fusing knowledge across individual models to yield a better single model. In this paper, we study the problem of merging individual models built on different training data sets to obtain a single model that performs well both across all data set domains and can generalize on out-of-domain data. We propose a dataless knowledge fusion method that merges models in their parameter space, guided by weights that minimize prediction differences between the merged model and the individual models. Over a battery of evaluation settings, we show that the proposed method significantly outperforms baselines such as Fisher-weighted averaging or model ensembling. Further, we find that our method is a promising alternative to multi-task learning that can preserve or sometimes improve over the individual models without access to the training data. Finally, model merging is more efficient than training a multi-task model, thus making it applicable to a wider set of scenarios.
title Dataless Knowledge Fusion by Merging Weights of Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2212.09849