@inproceedings{oneill-dutta-2023-self,
title = "Self-Distilled Quantization: Achieving High Compression Rates in Transformer-Based Language Models",
author = "O{'}Neill, James and
Dutta, Sourav",
editor = "Rogers, Anna and
Boyd-Graber, Jordan and
Okazaki, Naoaki",
booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.acl-short.114",
doi = "10.18653/v1/2023.acl-short.114",
pages = "1329--1339",
abstract = "We investigate the effects of post-training quantization and quantization-aware training on the generalization of Transformer language models. We present a new method called self-distilled quantization (SDQ) that minimizes accumulative quantization errors and outperforms baselines. We apply SDQ to multilingual models XLM-R$_{\text{Base}}$ and InfoXLM$_{\text{Base}}$ and demonstrate that both models can be reduced from 32-bit floating point weights to 8-bit integer weights while maintaining a high level of performance on the XGLUE benchmark. Our results also highlight the challenges of quantizing multilingual models, which must generalize to languages they were not fine-tuned on.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="oneill-dutta-2023-self">
<titleInfo>
<title>Self-Distilled Quantization: Achieving High Compression Rates in Transformer-Based Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">James</namePart>
<namePart type="family">O’Neill</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sourav</namePart>
<namePart type="family">Dutta</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>Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)</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>We investigate the effects of post-training quantization and quantization-aware training on the generalization of Transformer language models. We present a new method called self-distilled quantization (SDQ) that minimizes accumulative quantization errors and outperforms baselines. We apply SDQ to multilingual models XLM-R_\textBase and InfoXLM_\textBase and demonstrate that both models can be reduced from 32-bit floating point weights to 8-bit integer weights while maintaining a high level of performance on the XGLUE benchmark. Our results also highlight the challenges of quantizing multilingual models, which must generalize to languages they were not fine-tuned on.</abstract>
<identifier type="citekey">oneill-dutta-2023-self</identifier>
<identifier type="doi">10.18653/v1/2023.acl-short.114</identifier>
<location>
<url>https://aclanthology.org/2023.acl-short.114</url>
</location>
<part>
<date>2023-07</date>
<extent unit="page">
<start>1329</start>
<end>1339</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Self-Distilled Quantization: Achieving High Compression Rates in Transformer-Based Language Models
%A O’Neill, James
%A Dutta, Sourav
%Y Rogers, Anna
%Y Boyd-Graber, Jordan
%Y Okazaki, Naoaki
%S Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)
%D 2023
%8 July
%I Association for Computational Linguistics
%C Toronto, Canada
%F oneill-dutta-2023-self
%X We investigate the effects of post-training quantization and quantization-aware training on the generalization of Transformer language models. We present a new method called self-distilled quantization (SDQ) that minimizes accumulative quantization errors and outperforms baselines. We apply SDQ to multilingual models XLM-R_\textBase and InfoXLM_\textBase and demonstrate that both models can be reduced from 32-bit floating point weights to 8-bit integer weights while maintaining a high level of performance on the XGLUE benchmark. Our results also highlight the challenges of quantizing multilingual models, which must generalize to languages they were not fine-tuned on.
%R 10.18653/v1/2023.acl-short.114
%U https://aclanthology.org/2023.acl-short.114
%U https://doi.org/10.18653/v1/2023.acl-short.114
%P 1329-1339
Markdown (Informal)
[Self-Distilled Quantization: Achieving High Compression Rates in Transformer-Based Language Models](https://aclanthology.org/2023.acl-short.114) (O’Neill & Dutta, ACL 2023)
ACL