@inproceedings{zhao-etal-2024-probe-retrieve,
title = "Probe Then Retrieve and Reason: Distilling Probing and Reasoning Capabilities into Smaller Language Models",
author = "Zhao, Yichun and
Zhou, Shuheng and
Zhu, Huijia",
editor = "Calzolari, Nicoletta and
Kan, Min-Yen and
Hoste, Veronique and
Lenci, Alessandro and
Sakti, Sakriani and
Xue, Nianwen",
booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
month = may,
year = "2024",
address = "Torino, Italia",
publisher = "ELRA and ICCL",
url = "https://aclanthology.org/2024.lrec-main.1140",
pages = "13026--13032",
abstract = "Step-by-step reasoning methods, such as the Chain-of-Thought (CoT), have been demonstrated to be highly effective in harnessing the reasoning capabilities of Large Language Models (LLMs). Recent research efforts have sought to distill LLMs into Small Language Models (SLMs), with a significant focus on transferring the reasoning capabilities of LLMs to SLMs via CoT. However, the outcomes of CoT distillation are inadequate for knowledge-intensive reasoning tasks. This is because generating accurate rationales requires crucial factual knowledge, which SLMs struggle to retain due to their parameter constraints. We propose a retrieval-based CoT distillation framework, named Probe then Retrieve and Reason (PRR), which distills the question probing and reasoning capabilities from LLMs into SLMs. We train two complementary distilled SLMs, a probing model and a reasoning model, in tandem. When presented with a new question, the probing model first identifies the necessary knowledge to answer it, generating queries for retrieval. Subsequently, the reasoning model uses the retrieved knowledge to construct a step-by-step rationale for the answer. In knowledge-intensive reasoning tasks, such as StrategyQA and OpenbookQA, our distillation framework yields superior performance for SLMs compared to conventional methods, including simple CoT distillation and knowledge-augmented distillation using raw questions.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-etal-2024-probe-retrieve">
<titleInfo>
<title>Probe Then Retrieve and Reason: Distilling Probing and Reasoning Capabilities into Smaller Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yichun</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuheng</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Huijia</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nicoletta</namePart>
<namePart type="family">Calzolari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min-Yen</namePart>
<namePart type="family">Kan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Veronique</namePart>
<namePart type="family">Hoste</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alessandro</namePart>
<namePart type="family">Lenci</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sakriani</namePart>
<namePart type="family">Sakti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nianwen</namePart>
<namePart type="family">Xue</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA and ICCL</publisher>
<place>
<placeTerm type="text">Torino, Italia</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Step-by-step reasoning methods, such as the Chain-of-Thought (CoT), have been demonstrated to be highly effective in harnessing the reasoning capabilities of Large Language Models (LLMs). Recent research efforts have sought to distill LLMs into Small Language Models (SLMs), with a significant focus on transferring the reasoning capabilities of LLMs to SLMs via CoT. However, the outcomes of CoT distillation are inadequate for knowledge-intensive reasoning tasks. This is because generating accurate rationales requires crucial factual knowledge, which SLMs struggle to retain due to their parameter constraints. We propose a retrieval-based CoT distillation framework, named Probe then Retrieve and Reason (PRR), which distills the question probing and reasoning capabilities from LLMs into SLMs. We train two complementary distilled SLMs, a probing model and a reasoning model, in tandem. When presented with a new question, the probing model first identifies the necessary knowledge to answer it, generating queries for retrieval. Subsequently, the reasoning model uses the retrieved knowledge to construct a step-by-step rationale for the answer. In knowledge-intensive reasoning tasks, such as StrategyQA and OpenbookQA, our distillation framework yields superior performance for SLMs compared to conventional methods, including simple CoT distillation and knowledge-augmented distillation using raw questions.</abstract>
<identifier type="citekey">zhao-etal-2024-probe-retrieve</identifier>
<location>
<url>https://aclanthology.org/2024.lrec-main.1140</url>
</location>
<part>
<date>2024-05</date>
<extent unit="page">
<start>13026</start>
<end>13032</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Probe Then Retrieve and Reason: Distilling Probing and Reasoning Capabilities into Smaller Language Models
%A Zhao, Yichun
%A Zhou, Shuheng
%A Zhu, Huijia
%Y Calzolari, Nicoletta
%Y Kan, Min-Yen
%Y Hoste, Veronique
%Y Lenci, Alessandro
%Y Sakti, Sakriani
%Y Xue, Nianwen
%S Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)
%D 2024
%8 May
%I ELRA and ICCL
%C Torino, Italia
%F zhao-etal-2024-probe-retrieve
%X Step-by-step reasoning methods, such as the Chain-of-Thought (CoT), have been demonstrated to be highly effective in harnessing the reasoning capabilities of Large Language Models (LLMs). Recent research efforts have sought to distill LLMs into Small Language Models (SLMs), with a significant focus on transferring the reasoning capabilities of LLMs to SLMs via CoT. However, the outcomes of CoT distillation are inadequate for knowledge-intensive reasoning tasks. This is because generating accurate rationales requires crucial factual knowledge, which SLMs struggle to retain due to their parameter constraints. We propose a retrieval-based CoT distillation framework, named Probe then Retrieve and Reason (PRR), which distills the question probing and reasoning capabilities from LLMs into SLMs. We train two complementary distilled SLMs, a probing model and a reasoning model, in tandem. When presented with a new question, the probing model first identifies the necessary knowledge to answer it, generating queries for retrieval. Subsequently, the reasoning model uses the retrieved knowledge to construct a step-by-step rationale for the answer. In knowledge-intensive reasoning tasks, such as StrategyQA and OpenbookQA, our distillation framework yields superior performance for SLMs compared to conventional methods, including simple CoT distillation and knowledge-augmented distillation using raw questions.
%U https://aclanthology.org/2024.lrec-main.1140
%P 13026-13032
Markdown (Informal)
[Probe Then Retrieve and Reason: Distilling Probing and Reasoning Capabilities into Smaller Language Models](https://aclanthology.org/2024.lrec-main.1140) (Zhao et al., LREC-COLING 2024)
ACL