Big Reasoning with Small Models: Instruction Retrieval at Inference Time

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Alkiek, Kenan, Jurgens, David, Vydiswaran, Vinod
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866917188220223488
author Alkiek, Kenan
Jurgens, David
Vydiswaran, Vinod
author_facet Alkiek, Kenan
Jurgens, David
Vydiswaran, Vinod
contents Small language models (SLMs) enable low-cost, private, on-device inference, but they often fail on problems that require specialized domain knowledge or multi-step reasoning. Existing approaches for improving reasoning either rely on scale (e.g., chain-of-thought prompting), require task-specific training that limits reuse and generality (e.g., distillation), or retrieve unstructured information that still leaves the SLM to determine an appropriate reasoning strategy. We propose instruction retrieval, an inference-time intervention that augments an SLM with structured, reusable reasoning procedures rather than raw passages. We construct an Instruction Corpus by clustering similar training questions and using a teacher model to generate generalizable guides that pair domain background with explicit step-by-step procedures. At inference, the SLM retrieves the instructions most relevant to a given query and executes the associated procedures without any additional fine-tuning. Across three challenging domains: medicine, law, and mathematics, instruction retrieval yields consistent gains for models with at least 3B parameters, improving accuracy by 9.4%, 7.9%, and 5.1%, respectively, with the strongest 14B model surpassing GPT-4o's zero-shot performance on knowledge-intensive tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2510_13935
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Big Reasoning with Small Models: Instruction Retrieval at Inference Time
Alkiek, Kenan
Jurgens, David
Vydiswaran, Vinod
Computation and Language
Artificial Intelligence
Small language models (SLMs) enable low-cost, private, on-device inference, but they often fail on problems that require specialized domain knowledge or multi-step reasoning. Existing approaches for improving reasoning either rely on scale (e.g., chain-of-thought prompting), require task-specific training that limits reuse and generality (e.g., distillation), or retrieve unstructured information that still leaves the SLM to determine an appropriate reasoning strategy. We propose instruction retrieval, an inference-time intervention that augments an SLM with structured, reusable reasoning procedures rather than raw passages. We construct an Instruction Corpus by clustering similar training questions and using a teacher model to generate generalizable guides that pair domain background with explicit step-by-step procedures. At inference, the SLM retrieves the instructions most relevant to a given query and executes the associated procedures without any additional fine-tuning. Across three challenging domains: medicine, law, and mathematics, instruction retrieval yields consistent gains for models with at least 3B parameters, improving accuracy by 9.4%, 7.9%, and 5.1%, respectively, with the strongest 14B model surpassing GPT-4o's zero-shot performance on knowledge-intensive tasks.
title Big Reasoning with Small Models: Instruction Retrieval at Inference Time
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2510.13935