Cranky Man's Guide to LoRA & QLoRA: Personal Lessons from a Thousand LLM Fine-Tuning Fails

This book is presented as a highly personal, opinionated, and practical guide to fine-tuning Large Language Models (LLMs), specifically using LoRA and QLoRA techniques. Written in a distinctly cranky, humorous, and often cynical voice by an experienced programmer and LLM tool builder ("Cranky Man"), it aims to cut through AI hype and technical jargon to provide hard-won lessons learned from extensive experimentation and failure.

The author will guide you through the minefield of AI terms and concepts and you will emerge at the other side, preferably still with your sanity intact.

• A solid grounding in the theory of Large Language Models, without getting bogged down in the details and math. You'll learn about the different types of models, quantization, how to choose the right one for the job, and how to train it.
• What all of the rank, alpha, batch_size or gradient accumulation actually do. In a plain language.
• You'll discover the tricks of the trade, the secrets of data collection, preparation and augmentation.
• You'll learn how to set up and use the hardware and software; even how to build your own multi-GPU AI supercomputer.
• How to read the tea leaves: Demystify the training process by learning to interpret those inscrutable loss graphs, and tell when your model is actually learning, when it's just memorizing the training data, and when it's blissfully and expensively dead.
• How to sidestep the pitfalls: Benefit from Cranky Man's hard-won experience and avoid the most common CUDA errors, Python tantrums and soul-crushing weight of unmet expectations.

The author, F.P. HAM aka Cranky Man, is a real-world senior programmer. He didn't just use the training tools, he contributed and built many of them! From the rage-coded Training PRO, to the VirtualLora, Twinbook, Playground or Storycrafter.

Cranky Man has been there, done that, got the T-shirt, the GPU burn marks and the electricity bills to prove it!

This book is the result of thousands of experiments, the successes, the near misses, the total disasters. It is the story of one man's obsessive quest to create the ultimate AI model.

So, there you have it. The definitive guide to training your own Large Language Model. If you're serious about getting into AI, this is the book for you. It's the one book that Cranky Man wished he'd had when he started out. He says it would have saved him a small country GDP-worth in electricity.