Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Klein, Aaron, Golebiowski, Jacek, Ma, Xingchen, Perrone, Valerio, Archambeau, Cedric
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2405.02267
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914923293966336
author Klein, Aaron
Golebiowski, Jacek
Ma, Xingchen
Perrone, Valerio
Archambeau, Cedric
author_facet Klein, Aaron
Golebiowski, Jacek
Ma, Xingchen
Perrone, Valerio
Archambeau, Cedric
contents Pre-trained language models (PLM), for example BERT or RoBERTa, mark the state-of-the-art for natural language understanding task when fine-tuned on labeled data. However, their large size poses challenges in deploying them for inference in real-world applications, due to significant GPU memory requirements and high inference latency. This paper explores neural architecture search (NAS) for structural pruning to find sub-parts of the fine-tuned network that optimally trade-off efficiency, for example in terms of model size or latency, and generalization performance. We also show how we can utilize more recently developed two-stage weight-sharing NAS approaches in this setting to accelerate the search process. Unlike traditional pruning methods with fixed thresholds, we propose to adopt a multi-objective approach that identifies the Pareto optimal set of sub-networks, allowing for a more flexible and automated compression process.
format Preprint
id arxiv_https___arxiv_org_abs_2405_02267
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Structural Pruning of Pre-trained Language Models via Neural Architecture Search
Klein, Aaron
Golebiowski, Jacek
Ma, Xingchen
Perrone, Valerio
Archambeau, Cedric
Machine Learning
Computation and Language
Pre-trained language models (PLM), for example BERT or RoBERTa, mark the state-of-the-art for natural language understanding task when fine-tuned on labeled data. However, their large size poses challenges in deploying them for inference in real-world applications, due to significant GPU memory requirements and high inference latency. This paper explores neural architecture search (NAS) for structural pruning to find sub-parts of the fine-tuned network that optimally trade-off efficiency, for example in terms of model size or latency, and generalization performance. We also show how we can utilize more recently developed two-stage weight-sharing NAS approaches in this setting to accelerate the search process. Unlike traditional pruning methods with fixed thresholds, we propose to adopt a multi-objective approach that identifies the Pareto optimal set of sub-networks, allowing for a more flexible and automated compression process.
title Structural Pruning of Pre-trained Language Models via Neural Architecture Search
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2405.02267