# LLM Fine-Tuning

> Eight posts taking an LLM from should-I-even-fine-tune-this to running in production, each with video and audio.

- Author: Nyghtowl (Melanie Warrick)
- URL: https://nyghtowl.com/fine-tuning/
- Published: 2026-08-17
- Site index for agents: https://nyghtowl.com/llms.txt


[Start the guide](https://nyghtowl.com/fine-tuning/#start-here)



Eight posts on fine-tuning an LLM, written over the back half of 2025 while I was doing it. This page puts them in the order you'd want to read them, since the blog lists them newest first and you end up landing on deployment before you know what fine-tuning is.

It covers the whole arc: deciding whether you need to fine-tune at all, getting a GPU to cooperate, quantization, the tradeoffs nobody warns you about, evaluating whether the thing actually works, and shipping it. Every post has an audio version, most have a video, and a few have a two-minute clip if you just want the gist.

<div style="position: relative; aspect-ratio: 16 / 9; overflow: hidden; max-width: 560px; margin: 16px 0; border-radius: 6px;">
  <iframe src="https://www.youtube.com/embed/rlODORkycTI" title="Building with LLMs: From Chat to Custom AI" style="position: absolute; top: 0; left: 0; width: 100%; height: 100%; border: 0;" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe>
</div>

## Start here

**1. [Building with LLMs: From Chat to Custom AI](https://nyghtowl.com/posts/2025/11/building-with-llms-from-chat-to-custom-ai/)**
When to prompt, when to do RAG, and when fine-tuning is actually the right answer. Worth reading before you spend money on a GPU, since sometimes the answer is that you don't need to fine-tune at all.
[Video](https://youtu.be/rlODORkycTI) · [Audio](https://youtu.be/N5qllVP6oCI) · [Clip](https://youtu.be/5jrpG1ehFhY)

**2. [What Is LLM Fine-Tuning?](https://nyghtowl.com/posts/2025/10/what-is-llm-fine-tuning/)**
What's actually happening under the hood, explained without the math, and how a general model becomes a domain expert.
[Video](https://youtu.be/YxulHbFh5UA) · [Audio](https://youtu.be/S-BKwd-4jjk)

## Getting set up

**3. [So You Want to Fine-Tune a Model (and You Finally Got a GPU!)](https://nyghtowl.com/posts/2025/12/so-you-want-to-fine-tune-a-model/)**
Hardware, environment, and the setup tips I wish someone had handed me before I lost a weekend to them.
[Audio](https://youtu.be/k4D4oES_dHE) · [Clip](https://youtu.be/Wy1S7Os4Nmc)

**4. [How to Run Big Models on Small GPUs](https://nyghtowl.com/posts/2025/11/how-to-run-big-models-on-small-gpus/)**
All about quantization: how to fit a model that doesn't fit, and what you give up when you do.
[Video](https://youtu.be/9q9cOpMHiyQ) · [Audio](https://youtu.be/4VGvm7dN6WU) · [Clip](https://youtu.be/-f5K3I0Dd4k)

## Tuning it

**5. [LLM Fine-Tuning & Performance Tug-of-War](https://nyghtowl.com/posts/2025/10/llm-fine-tuning-performance-tug-of-war/)**
Every knob you turn costs you something somewhere else, so this maps out the tradeoffs.
[Video](https://youtu.be/p6z5-CbeFic) · [Audio](https://youtu.be/OyRJjiCYKmM)

**6. [LLM Fine-Tuning Optimization Part 2: Achieving Stability](https://nyghtowl.com/posts/2025/11/llm-fine-tuning-optimization-part-2/)**
Part two of the tug-of-war, on getting training runs that don't fall over.
[Video](https://youtu.be/WgLB0uCUBgE) · [Audio](https://youtu.be/q9DUw2tnw8Y)

## Shipping it

**7. [Your Fine-Tuned LLM Model Isn't Ready Yet](https://nyghtowl.com/posts/2025/12/your-fine-tuned-llm-model-isnt-ready/)**
How to evaluate it before you let anyone near it. A loss curve that looks nice is not evidence that the thing works.
[Video](https://youtu.be/4Z0PvxWta2I) · [Audio](https://youtu.be/dp57oO5p4LI) · [Clip](https://youtu.be/ZRUX95pGrzA)

**8. [Opening Night: Launching a Fine-Tuned LLM to Production](https://nyghtowl.com/posts/2025/12/opening-night-launching-a-fine-tuned-llm/)**
Turning a checkpoint into a service real people can hit, including what happens when the GPU runs out of memory on New Year's Eve.
[Video](https://youtu.be/Drd_1fekEqU) · [Audio](https://youtu.be/ISJqjCcEjnI)

## Working on your own model?

[I can help you choose, fine-tune, evaluate, and deploy it](https://nyghtowl.com/consulting/#models), including figuring out whether fine-tuning is the right approach.

## Elsewhere

The full set also lives on the [blog series page](https://nyghtowl.com/series/llm-fine-tuning/) in publication order, and the videos are on [YouTube](https://www.youtube.com/@nyghtowlYT). LinkedIn is the easiest place to reach me if you're working through this and something doesn't land.