@inproceedings{yam-paek-2024-teaching,
title = "Teaching Tiny Minds: Exploring Methods to Enhance Knowledge Distillation for Small Language Models",
author = "Yam, Hong Meng and
Paek, Nathan",
editor = "Hu, Michael Y. and
Mueller, Aaron and
Ross, Candace and
Williams, Adina and
Linzen, Tal and
Zhuang, Chengxu and
Choshen, Leshem and
Cotterell, Ryan and
Warstadt, Alex and
Wilcox, Ethan Gotlieb",
booktitle = "The 2nd BabyLM Challenge at the 28th Conference on Computational Natural Language Learning",
month = nov,
year = "2024",
address = "Miami, FL, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.conll-babylm.27/",
pages = "302--307",
abstract = "In this paper, we build off of the success of the previous BabyLM challenge winner`s model, BabyLlama, to explore various methods of enhancing knowledge distillation for small language models. Our main focus is on investigating how small a language model can be while still maintaining competitive performance. We experiment with three main approaches: (1) DistilledGPT-44M, which uses smaller teacher models and a more compact student model compared to BabyLlama; (2) ContrastiveLlama-58M, which incorporates contrastive loss into the knowledge distillation process; and (3) MaskedAdversarialLlama-58M, incorporates adversarial loss into the knowledge distillation process. Using the 10M-word dataset from the BabyLM challenge`s strict-small track, we evaluate our models on the BLiMP, EWoK, and GLUE benchmarks. Our results show that effective knowledge distillation can still be achieved with significantly smaller teacher and student models. In particular, our model DistilledGPT-44M is able to achieve better performance than one of last year`s winning entries, LTG-BERT, while achieving similar performance but cutting training time by around 70{\%} and parameters by around 25{\%} compared to the other winning entry, BabyLlama."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yam-paek-2024-teaching">
<titleInfo>
<title>Teaching Tiny Minds: Exploring Methods to Enhance Knowledge Distillation for Small Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hong</namePart>
<namePart type="given">Meng</namePart>
<namePart type="family">Yam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nathan</namePart>
<namePart type="family">Paek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>The 2nd BabyLM Challenge at the 28th Conference on Computational Natural Language Learning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Michael</namePart>
<namePart type="given">Y</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Candace</namePart>
<namePart type="family">Ross</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Adina</namePart>
<namePart type="family">Williams</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tal</namePart>
<namePart type="family">Linzen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chengxu</namePart>
<namePart type="family">Zhuang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leshem</namePart>
<namePart type="family">Choshen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ryan</namePart>
<namePart type="family">Cotterell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alex</namePart>
<namePart type="family">Warstadt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ethan</namePart>
<namePart type="given">Gotlieb</namePart>
<namePart type="family">Wilcox</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Miami, FL, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In this paper, we build off of the success of the previous BabyLM challenge winner‘s model, BabyLlama, to explore various methods of enhancing knowledge distillation for small language models. Our main focus is on investigating how small a language model can be while still maintaining competitive performance. We experiment with three main approaches: (1) DistilledGPT-44M, which uses smaller teacher models and a more compact student model compared to BabyLlama; (2) ContrastiveLlama-58M, which incorporates contrastive loss into the knowledge distillation process; and (3) MaskedAdversarialLlama-58M, incorporates adversarial loss into the knowledge distillation process. Using the 10M-word dataset from the BabyLM challenge‘s strict-small track, we evaluate our models on the BLiMP, EWoK, and GLUE benchmarks. Our results show that effective knowledge distillation can still be achieved with significantly smaller teacher and student models. In particular, our model DistilledGPT-44M is able to achieve better performance than one of last year‘s winning entries, LTG-BERT, while achieving similar performance but cutting training time by around 70% and parameters by around 25% compared to the other winning entry, BabyLlama.</abstract>
<identifier type="citekey">yam-paek-2024-teaching</identifier>
<location>
<url>https://aclanthology.org/2024.conll-babylm.27/</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>302</start>
<end>307</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Teaching Tiny Minds: Exploring Methods to Enhance Knowledge Distillation for Small Language Models
%A Yam, Hong Meng
%A Paek, Nathan
%Y Hu, Michael Y.
%Y Mueller, Aaron
%Y Ross, Candace
%Y Williams, Adina
%Y Linzen, Tal
%Y Zhuang, Chengxu
%Y Choshen, Leshem
%Y Cotterell, Ryan
%Y Warstadt, Alex
%Y Wilcox, Ethan Gotlieb
%S The 2nd BabyLM Challenge at the 28th Conference on Computational Natural Language Learning
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, FL, USA
%F yam-paek-2024-teaching
%X In this paper, we build off of the success of the previous BabyLM challenge winner‘s model, BabyLlama, to explore various methods of enhancing knowledge distillation for small language models. Our main focus is on investigating how small a language model can be while still maintaining competitive performance. We experiment with three main approaches: (1) DistilledGPT-44M, which uses smaller teacher models and a more compact student model compared to BabyLlama; (2) ContrastiveLlama-58M, which incorporates contrastive loss into the knowledge distillation process; and (3) MaskedAdversarialLlama-58M, incorporates adversarial loss into the knowledge distillation process. Using the 10M-word dataset from the BabyLM challenge‘s strict-small track, we evaluate our models on the BLiMP, EWoK, and GLUE benchmarks. Our results show that effective knowledge distillation can still be achieved with significantly smaller teacher and student models. In particular, our model DistilledGPT-44M is able to achieve better performance than one of last year‘s winning entries, LTG-BERT, while achieving similar performance but cutting training time by around 70% and parameters by around 25% compared to the other winning entry, BabyLlama.
%U https://aclanthology.org/2024.conll-babylm.27/
%P 302-307
Markdown (Informal)
[Teaching Tiny Minds: Exploring Methods to Enhance Knowledge Distillation for Small Language Models](https://aclanthology.org/2024.conll-babylm.27/) (Yam & Paek, CoNLL-BabyLM 2024)
ACL