@inproceedings{horawalavithana-etal-2024-scitune,
title = "{SCITUNE}: Aligning Large Language Models with Human-Curated Scientific Multimodal Instructions",
author = "Horawalavithana, Sameera and
Munikoti, Sai and
Stewart, Ian and
Kvinge, Henry and
Pazdernik, Karl",
editor = "Peled-Cohen, Lotem and
Calderon, Nitay and
Lissak, Shir and
Reichart, Roi",
booktitle = "Proceedings of the 1st Workshop on NLP for Science (NLP4Science)",
month = nov,
year = "2024",
address = "Miami, FL, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.nlp4science-1.7",
pages = "58--72",
abstract = "Instruction finetuning is a popular paradigm to align large language models (LLM) with human intent. Despite its popularity, this idea is less explored in improving LLMs to align existing foundation models with scientific disciplines, concepts and goals. In this work, we present SciTune as a tuning framework to improve the ability of LLMs to follow multimodal instructions generated from scientific publications. To test our methodology, we train a large multimodal model LLaMA-SciTune that connects a vision encoder and LLM for science-focused visual and language understanding. LLaMA-SciTune significantly outperforms the state-of-the-art models in the generated figure types and captions in SciCap and VisText benchmarks. In comparison to the models that are finetuned with synthetic data only, LLaMA-SciTune surpasses human performance on average and in many sub-categories on the ScienceQA benchmark. Our results demonstrate that human-generated scientific multimodal instructions remain highly valuable in tuning LLMs to perform well on science tasks, despite their lower volume and relative scarcity compared to synthetic data.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="horawalavithana-etal-2024-scitune">
<titleInfo>
<title>SCITUNE: Aligning Large Language Models with Human-Curated Scientific Multimodal Instructions</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sameera</namePart>
<namePart type="family">Horawalavithana</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sai</namePart>
<namePart type="family">Munikoti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ian</namePart>
<namePart type="family">Stewart</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Henry</namePart>
<namePart type="family">Kvinge</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Karl</namePart>
<namePart type="family">Pazdernik</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>Proceedings of the 1st Workshop on NLP for Science (NLP4Science)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lotem</namePart>
<namePart type="family">Peled-Cohen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nitay</namePart>
<namePart type="family">Calderon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shir</namePart>
<namePart type="family">Lissak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roi</namePart>
<namePart type="family">Reichart</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>Instruction finetuning is a popular paradigm to align large language models (LLM) with human intent. Despite its popularity, this idea is less explored in improving LLMs to align existing foundation models with scientific disciplines, concepts and goals. In this work, we present SciTune as a tuning framework to improve the ability of LLMs to follow multimodal instructions generated from scientific publications. To test our methodology, we train a large multimodal model LLaMA-SciTune that connects a vision encoder and LLM for science-focused visual and language understanding. LLaMA-SciTune significantly outperforms the state-of-the-art models in the generated figure types and captions in SciCap and VisText benchmarks. In comparison to the models that are finetuned with synthetic data only, LLaMA-SciTune surpasses human performance on average and in many sub-categories on the ScienceQA benchmark. Our results demonstrate that human-generated scientific multimodal instructions remain highly valuable in tuning LLMs to perform well on science tasks, despite their lower volume and relative scarcity compared to synthetic data.</abstract>
<identifier type="citekey">horawalavithana-etal-2024-scitune</identifier>
<location>
<url>https://aclanthology.org/2024.nlp4science-1.7</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>58</start>
<end>72</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SCITUNE: Aligning Large Language Models with Human-Curated Scientific Multimodal Instructions
%A Horawalavithana, Sameera
%A Munikoti, Sai
%A Stewart, Ian
%A Kvinge, Henry
%A Pazdernik, Karl
%Y Peled-Cohen, Lotem
%Y Calderon, Nitay
%Y Lissak, Shir
%Y Reichart, Roi
%S Proceedings of the 1st Workshop on NLP for Science (NLP4Science)
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, FL, USA
%F horawalavithana-etal-2024-scitune
%X Instruction finetuning is a popular paradigm to align large language models (LLM) with human intent. Despite its popularity, this idea is less explored in improving LLMs to align existing foundation models with scientific disciplines, concepts and goals. In this work, we present SciTune as a tuning framework to improve the ability of LLMs to follow multimodal instructions generated from scientific publications. To test our methodology, we train a large multimodal model LLaMA-SciTune that connects a vision encoder and LLM for science-focused visual and language understanding. LLaMA-SciTune significantly outperforms the state-of-the-art models in the generated figure types and captions in SciCap and VisText benchmarks. In comparison to the models that are finetuned with synthetic data only, LLaMA-SciTune surpasses human performance on average and in many sub-categories on the ScienceQA benchmark. Our results demonstrate that human-generated scientific multimodal instructions remain highly valuable in tuning LLMs to perform well on science tasks, despite their lower volume and relative scarcity compared to synthetic data.
%U https://aclanthology.org/2024.nlp4science-1.7
%P 58-72
Markdown (Informal)
[SCITUNE: Aligning Large Language Models with Human-Curated Scientific Multimodal Instructions](https://aclanthology.org/2024.nlp4science-1.7) (Horawalavithana et al., NLP4Science 2024)
ACL