@inproceedings{thangarasa-etal-2024-mediswift,
title = "{M}edi{S}wift: Efficient Sparse Pre-trained Biomedical Language Models",
author = "Thangarasa, Vithursan and
Salem, Mahmoud and
Saxena, Shreyas and
Leong, Chen-Yu and
Hestness, Joel and
Lie, Sean",
editor = "Ku, Lun-Wei and
Martins, Andre and
Srikumar, Vivek",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2024",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-acl.14/",
doi = "10.18653/v1/2024.findings-acl.14",
pages = "214--230",
abstract = "Large language models (LLMs) are typically trained on general source data forvarious domains, but a recent surge in domain-specific LLMs has shown theirpotential to outperform general-purpose models in domain-specific tasks (e.g.,biomedicine). Although domain-specific pre-training enhances efficiency andleads to smaller models, the computational costs of training these LLMs remainhigh, posing budgeting challenges. We introduce MediSwift, a suite of biomedicalLMs that leverage sparse pre-training on domain-specific biomedical text data.By inducing up to 75{\%} weight sparsity during the pre-training phase, MediSwiftachieves a 2-2.5x reduction in training FLOPs. Notably, all sparse pre-trainingwas performed on the Cerebras CS-2 system, which is specifically designed torealize the acceleration benefits from unstructured weight sparsity, therebysignificantly enhancing the efficiency of the MediSwift models. Throughsubsequent dense fine-tuning and strategic soft prompting, MediSwift modelsoutperform existing LLMs up to 7B parameters on biomedical tasks, setting newbenchmarks w.r.t efficiency-accuracy on tasks such as PubMedQA. Our results showthat sparse pre-training, along with dense fine-tuning and soft prompting,offers an effective method for creating high-performing, computationallyefficient models in specialized domains."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="thangarasa-etal-2024-mediswift">
<titleInfo>
<title>MediSwift: Efficient Sparse Pre-trained Biomedical Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vithursan</namePart>
<namePart type="family">Thangarasa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mahmoud</namePart>
<namePart type="family">Salem</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shreyas</namePart>
<namePart type="family">Saxena</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chen-Yu</namePart>
<namePart type="family">Leong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joel</namePart>
<namePart type="family">Hestness</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sean</namePart>
<namePart type="family">Lie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-08</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lun-Wei</namePart>
<namePart type="family">Ku</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andre</namePart>
<namePart type="family">Martins</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vivek</namePart>
<namePart type="family">Srikumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Bangkok, Thailand</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Large language models (LLMs) are typically trained on general source data forvarious domains, but a recent surge in domain-specific LLMs has shown theirpotential to outperform general-purpose models in domain-specific tasks (e.g.,biomedicine). Although domain-specific pre-training enhances efficiency andleads to smaller models, the computational costs of training these LLMs remainhigh, posing budgeting challenges. We introduce MediSwift, a suite of biomedicalLMs that leverage sparse pre-training on domain-specific biomedical text data.By inducing up to 75% weight sparsity during the pre-training phase, MediSwiftachieves a 2-2.5x reduction in training FLOPs. Notably, all sparse pre-trainingwas performed on the Cerebras CS-2 system, which is specifically designed torealize the acceleration benefits from unstructured weight sparsity, therebysignificantly enhancing the efficiency of the MediSwift models. Throughsubsequent dense fine-tuning and strategic soft prompting, MediSwift modelsoutperform existing LLMs up to 7B parameters on biomedical tasks, setting newbenchmarks w.r.t efficiency-accuracy on tasks such as PubMedQA. Our results showthat sparse pre-training, along with dense fine-tuning and soft prompting,offers an effective method for creating high-performing, computationallyefficient models in specialized domains.</abstract>
<identifier type="citekey">thangarasa-etal-2024-mediswift</identifier>
<identifier type="doi">10.18653/v1/2024.findings-acl.14</identifier>
<location>
<url>https://aclanthology.org/2024.findings-acl.14/</url>
</location>
<part>
<date>2024-08</date>
<extent unit="page">
<start>214</start>
<end>230</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MediSwift: Efficient Sparse Pre-trained Biomedical Language Models
%A Thangarasa, Vithursan
%A Salem, Mahmoud
%A Saxena, Shreyas
%A Leong, Chen-Yu
%A Hestness, Joel
%A Lie, Sean
%Y Ku, Lun-Wei
%Y Martins, Andre
%Y Srikumar, Vivek
%S Findings of the Association for Computational Linguistics: ACL 2024
%D 2024
%8 August
%I Association for Computational Linguistics
%C Bangkok, Thailand
%F thangarasa-etal-2024-mediswift
%X Large language models (LLMs) are typically trained on general source data forvarious domains, but a recent surge in domain-specific LLMs has shown theirpotential to outperform general-purpose models in domain-specific tasks (e.g.,biomedicine). Although domain-specific pre-training enhances efficiency andleads to smaller models, the computational costs of training these LLMs remainhigh, posing budgeting challenges. We introduce MediSwift, a suite of biomedicalLMs that leverage sparse pre-training on domain-specific biomedical text data.By inducing up to 75% weight sparsity during the pre-training phase, MediSwiftachieves a 2-2.5x reduction in training FLOPs. Notably, all sparse pre-trainingwas performed on the Cerebras CS-2 system, which is specifically designed torealize the acceleration benefits from unstructured weight sparsity, therebysignificantly enhancing the efficiency of the MediSwift models. Throughsubsequent dense fine-tuning and strategic soft prompting, MediSwift modelsoutperform existing LLMs up to 7B parameters on biomedical tasks, setting newbenchmarks w.r.t efficiency-accuracy on tasks such as PubMedQA. Our results showthat sparse pre-training, along with dense fine-tuning and soft prompting,offers an effective method for creating high-performing, computationallyefficient models in specialized domains.
%R 10.18653/v1/2024.findings-acl.14
%U https://aclanthology.org/2024.findings-acl.14/
%U https://doi.org/10.18653/v1/2024.findings-acl.14
%P 214-230
Markdown (Informal)
[MediSwift: Efficient Sparse Pre-trained Biomedical Language Models](https://aclanthology.org/2024.findings-acl.14/) (Thangarasa et al., Findings 2024)
ACL