@inproceedings{dasgupta-etal-2023-cost,
title = "Cost-effective Distillation of Large Language Models",
author = "Dasgupta, Sayantan and
Cohn, Trevor and
Baldwin, Timothy",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2023",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.findings-acl.463",
doi = "10.18653/v1/2023.findings-acl.463",
pages = "7346--7354",
abstract = "Knowledge distillation (KD) involves training a small {``}student{''} model to replicate the strong performance of a high-capacity {``}teacher{''} model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Several existing approaches require pretraining of the teacher on task-specific datasets, which can be costly for large and unstable for small datasets. Here we propose an approach for improving KD through a novel distillation loss agnostic to the task and model architecture. We successfully apply our method to the distillation of the BERT-base and achieve highly competitive results from the distilled student across a range of GLUE tasks, especially for tasks with smaller datasets.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dasgupta-etal-2023-cost">
<titleInfo>
<title>Cost-effective Distillation of Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sayantan</namePart>
<namePart type="family">Dasgupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Trevor</namePart>
<namePart type="family">Cohn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Timothy</namePart>
<namePart type="family">Baldwin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2023</title>
</titleInfo>
<name type="personal">
<namePart type="given">Anna</namePart>
<namePart type="family">Rogers</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jordan</namePart>
<namePart type="family">Boyd-Graber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoaki</namePart>
<namePart type="family">Okazaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Knowledge distillation (KD) involves training a small “student” model to replicate the strong performance of a high-capacity “teacher” model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Several existing approaches require pretraining of the teacher on task-specific datasets, which can be costly for large and unstable for small datasets. Here we propose an approach for improving KD through a novel distillation loss agnostic to the task and model architecture. We successfully apply our method to the distillation of the BERT-base and achieve highly competitive results from the distilled student across a range of GLUE tasks, especially for tasks with smaller datasets.</abstract>
<identifier type="citekey">dasgupta-etal-2023-cost</identifier>
<identifier type="doi">10.18653/v1/2023.findings-acl.463</identifier>
<location>
<url>https://aclanthology.org/2023.findings-acl.463</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>7346</start>
<end>7354</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Cost-effective Distillation of Large Language Models
%A Dasgupta, Sayantan
%A Cohn, Trevor
%A Baldwin, Timothy
%Y Rogers, Anna
%Y Boyd-Graber, Jordan
%Y Okazaki, Naoaki
%S Findings of the Association for Computational Linguistics: ACL 2023
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F dasgupta-etal-2023-cost
%X Knowledge distillation (KD) involves training a small “student” model to replicate the strong performance of a high-capacity “teacher” model, enabling efficient deployment in resource-constrained settings. Top-performing methods tend to be task- or architecture-specific and lack generalizability. Several existing approaches require pretraining of the teacher on task-specific datasets, which can be costly for large and unstable for small datasets. Here we propose an approach for improving KD through a novel distillation loss agnostic to the task and model architecture. We successfully apply our method to the distillation of the BERT-base and achieve highly competitive results from the distilled student across a range of GLUE tasks, especially for tasks with smaller datasets.
%R 10.18653/v1/2023.findings-acl.463
%U https://aclanthology.org/2023.findings-acl.463
%U https://doi.org/10.18653/v1/2023.findings-acl.463
%P 7346-7354
Markdown (Informal)
[Cost-effective Distillation of Large Language Models](https://aclanthology.org/2023.findings-acl.463) (Dasgupta et al., Findings 2023)
ACL
- Sayantan Dasgupta, Trevor Cohn, and Timothy Baldwin. 2023. Cost-effective Distillation of Large Language Models. In Findings of the Association for Computational Linguistics: ACL 2023, pages 7346–7354, Toronto, Canada. Association for Computational Linguistics.