Fine-Tuning AI

Chapter 15. Serving the Expert

A Note for Early Release Readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 15th chapter of the final book. Please note that the GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

You now have a private expert model that you can run! Over the past 14 chapters, you took a base Gemma 4 model, curated a medical Q&A dataset, trained a LoRA adapter, aligned it with DPO, merged it, and compressed it into a single GGUF file that loads on a laptop and still answers medical questions well. That file is the payoff of the whole book. But a file on disk is not a product. Nothing in your existing software can call it, no user can reach it, and it does nothing at all until a process loads it into memory and listens for requests.

That is just as true for the image models you built earlier. Back in Chapters 12 and 13, you left the language world entirely and fine-tuned diffusion models, teaching one a specific subject with DreamBooth and another a specific artistic style with a style LoRA.

All these models have the same problem: a trained checkpoint is not a service until something serves it. That is what this chapter is going to solve.

The serving principles ...