Domain-Specific Small Language Models

Building an AI model for a specific domain brings a different set of engineering questions: how should it be adapted, evaluated, optimized, and run where hardware or cost is limited? Domain-Specific Small Language Models by Guglielmo Iozzia addresses that practical path, from model tuning to local deployment.

Written for AI engineers familiar with Python, this book links the foundations of small language models to the choices involved in putting them to work. Its examples range from Python code generation to protein structures, showing the breadth of specialized applications explored.

From Domain Data to a Tuned Model

The early chapters introduce small language models and Transformers, then examine how to prepare data and adapt pretrained models. Fine-tuning, parameter-efficient approaches such as LoRA, and retrieval-augmented generation are considered alongside the question of when to use RAG and when to fine-tune.

Make Inference and Optimization Practical

Readers move on to inference, evaluation, and cost considerations, followed by ONNX and quantization techniques. The coverage includes 8-bit and 4-bit methods, advanced quantization approaches, profiling, and optimization—topics that matter when a model has to fit a particular production environment rather than an idealized one.

Real-World Applications, Local Deployment

Use cases include generating Python code and protein structures. Later chapters turn to deployment and serving, running models on a laptop, building end-to-end LLM applications, and exploring advanced application components and test-time compute.

For Engineers Working with Specialized AI

This is a technical, applied guide for Python-familiar AI engineers who want to understand the steps between a pretrained model and a domain-focused system. Its progression through data, adaptation, runtime, optimization, and deployment gives readers a connected view of the work—not just a tour of model terminology.

If your interest is in making language models more focused and practical under real hardware and deployment constraints, Iozzia’s book offers a structured place to explore the engineering choices involved.