@inproceedings{zheng-etal-2025-lightweight,
title = "Lightweight Safety Guardrails Using Fine-tuned {BERT} Embeddings",
author = "Zheng, Aaron and
Rana, Mansi and
Stolcke, Andreas",
editor = "Rambow, Owen and
Wanner, Leo and
Apidianaki, Marianna and
Al-Khalifa, Hend and
Eugenio, Barbara Di and
Schockaert, Steven and
Darwish, Kareem and
Agarwal, Apoorv",
booktitle = "Proceedings of the 31st International Conference on Computational Linguistics: Industry Track",
month = jan,
year = "2025",
address = "Abu Dhabi, UAE",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.coling-industry.58/",
pages = "689--696",
abstract = "With the recent proliferation of large language models (LLMs), enterprises have been able to rapidly develop proof-of-concepts and prototypes. As a result, there is a growing need to implement robust guardrails that monitor, quantize and control an LLM`s behavior, ensuring that the use is reliable, safe, accurate and also aligned with the users' expectations. Previous approaches for filtering out inappropriate user prompts or system outputs, such as LlamaGuard and OpenAI`s MOD API, have achieved significant success by fine-tuning existing LLMs. However, using fine-tuned LLMs as guardrails introduces increased latency and higher maintenance costs, which may not be practical or scalable for cost-efficient deployments. We take a different approach, focusing on fine-tuning a lightweight architecture: Sentence-BERT. This method reduces the model size from LlamaGuard`s 7 billion parameters to approximately 67 million, while maintaining comparable performance on the AEGIS safety benchmark."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zheng-etal-2025-lightweight">
<titleInfo>
<title>Lightweight Safety Guardrails Using Fine-tuned BERT Embeddings</title>
</titleInfo>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mansi</namePart>
<namePart type="family">Rana</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andreas</namePart>
<namePart type="family">Stolcke</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-01</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 31st International Conference on Computational Linguistics: Industry Track</title>
</titleInfo>
<name type="personal">
<namePart type="given">Owen</namePart>
<namePart type="family">Rambow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leo</namePart>
<namePart type="family">Wanner</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marianna</namePart>
<namePart type="family">Apidianaki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hend</namePart>
<namePart type="family">Al-Khalifa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Barbara</namePart>
<namePart type="given">Di</namePart>
<namePart type="family">Eugenio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Schockaert</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kareem</namePart>
<namePart type="family">Darwish</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Apoorv</namePart>
<namePart type="family">Agarwal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Abu Dhabi, UAE</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>With the recent proliferation of large language models (LLMs), enterprises have been able to rapidly develop proof-of-concepts and prototypes. As a result, there is a growing need to implement robust guardrails that monitor, quantize and control an LLM‘s behavior, ensuring that the use is reliable, safe, accurate and also aligned with the users’ expectations. Previous approaches for filtering out inappropriate user prompts or system outputs, such as LlamaGuard and OpenAI‘s MOD API, have achieved significant success by fine-tuning existing LLMs. However, using fine-tuned LLMs as guardrails introduces increased latency and higher maintenance costs, which may not be practical or scalable for cost-efficient deployments. We take a different approach, focusing on fine-tuning a lightweight architecture: Sentence-BERT. This method reduces the model size from LlamaGuard‘s 7 billion parameters to approximately 67 million, while maintaining comparable performance on the AEGIS safety benchmark.</abstract>
<identifier type="citekey">zheng-etal-2025-lightweight</identifier>
<location>
<url>https://aclanthology.org/2025.coling-industry.58/</url>
</location>
<part>
<date>2025-01</date>
<extent unit="page">
<start>689</start>
<end>696</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Lightweight Safety Guardrails Using Fine-tuned BERT Embeddings
%A Zheng, Aaron
%A Rana, Mansi
%A Stolcke, Andreas
%Y Rambow, Owen
%Y Wanner, Leo
%Y Apidianaki, Marianna
%Y Al-Khalifa, Hend
%Y Eugenio, Barbara Di
%Y Schockaert, Steven
%Y Darwish, Kareem
%Y Agarwal, Apoorv
%S Proceedings of the 31st International Conference on Computational Linguistics: Industry Track
%D 2025
%8 January
%I Association for Computational Linguistics
%C Abu Dhabi, UAE
%F zheng-etal-2025-lightweight
%X With the recent proliferation of large language models (LLMs), enterprises have been able to rapidly develop proof-of-concepts and prototypes. As a result, there is a growing need to implement robust guardrails that monitor, quantize and control an LLM‘s behavior, ensuring that the use is reliable, safe, accurate and also aligned with the users’ expectations. Previous approaches for filtering out inappropriate user prompts or system outputs, such as LlamaGuard and OpenAI‘s MOD API, have achieved significant success by fine-tuning existing LLMs. However, using fine-tuned LLMs as guardrails introduces increased latency and higher maintenance costs, which may not be practical or scalable for cost-efficient deployments. We take a different approach, focusing on fine-tuning a lightweight architecture: Sentence-BERT. This method reduces the model size from LlamaGuard‘s 7 billion parameters to approximately 67 million, while maintaining comparable performance on the AEGIS safety benchmark.
%U https://aclanthology.org/2025.coling-industry.58/
%P 689-696
Markdown (Informal)
[Lightweight Safety Guardrails Using Fine-tuned BERT Embeddings](https://aclanthology.org/2025.coling-industry.58/) (Zheng et al., COLING 2025)
ACL