Chapter 09 · Sustainability & AISustainability & AI

Small language model (SLM)

Meaning statusEstablishedSource recordDirect document linkedWhy these are different

Definition

A language model small enough to run with modest compute — on a workstation, device or single accelerator — trading peak capability for efficiency and deployability.

References

arXiv (Microsoft)Gunasekar et al. — Textbooks Are All You Need (Phi-1, 2023)

This source provides part of the technical or institutional basis for the definition.

Communications of the ACM 2020Schwartz et al. — Green AI

This source supports the explanation of how the term is applied, measured or governed in practice.

Overview

What it means

SLMs pursue performance through training-data quality and design rather than sheer scale; Microsoft's Phi series popularised the approach with its 'Textbooks Are All You Need' recipe. They can run locally, keeping data in-house and avoiding hyperscale inference infrastructure.

How it is used

Use cases include on-device assistants, document triage inside secure environments, and lightweight analysis where a frontier model would be disproportionate — a pattern aligned with frugal-AI thinking.

Why it matters

SLMs are the practical counterweight to scale-at-all-costs AI: for many sustainability tasks they deliver most of the value at a fraction of the energy, cost and data-governance exposure.