Fine-tune open-source language models with LoRA, QLoRA, and Hugging Face tools

Fine-Tuning Large and Small Language Models walks practitioners through the complete pipeline for customizing open-source SLMs for domain-specific tasks. Written by Luca Massaron, a data scientist with 20+ years in data modelling and nearly a decade building NLP solutions with Transformer architectures, the book covers dataset preparation, synthetic data generation, base model selection from families including Gemma, Qwen, Phi, and Llama, and deployment on consumer-grade hardware.

The book frames fine-tuning against alternatives like retrieval-augmented generation and advanced prompting, helping readers determine when fine-tuning is the right approach. Hands-on coverage of parameter-efficient methods, specifically LoRA and QLoRA, shows how to configure Hugging Face PEFT, TRL, and bitsandbytes for training. Evaluation chapters address detecting whether fine-tuning improved target performance without degrading the model's broader capabilities.

Readers will also find:

• Practical case studies covering the end-to-end process from dataset preparation through model evaluation and production deployment
• Guidance on selecting base models from the Gemma, Qwen, Phi, and Llama families for specific use cases
• Techniques for generating synthetic training data where real domain-specific data is scarce or unavailable
• Configuration walkthroughs for Hugging Face PEFT, TRL, and bitsandbytes to run training on consumer hardware
• Final chapters extending fine-tuned SLMs toward autonomous agents and domain-specific production applications

Fine-Tuning Large and Small Language Models serves technical practitioners with programming and machine learning experience who want to move beyond off-the-shelf APIs. Data scientists, ML engineers, and AI developers building customized, cost-effective language models will gain the working knowledge to transform general-purpose SLMs into specialized, production-ready tools.