How to Train Your Model
A crash course in training local LLMs
The big models live in someone else’s data center. You rent them by the word, send them your data, and take whatever they were tuned to do. A small model on your own machine can instead be trained to do one job well.
This course covers the whole process on ordinary hardware: choosing a base model, preparing your data, fine-tuning it, and finding out whether it got better. Starting a training run is easy; knowing whether it worked is harder, and most of the course is about that.
Who it’s for
For people who want a model of their own: choosing one, feeding it data, fine-tuning it, and proving it got better.
What we cover
- 01
Hatching
What am I starting with?
- 02
Feeding
What do I teach it with?
- 03
Training
How does it actually learn?
- 04
Flying
Did it work, and can I use it?
What you’ll leave with
- A model of your own
- A small model fine-tuned on your data, running on your machine
- An evaluation set
- A way to tell whether a model got better, not just different
- A data pipeline
- A repeatable path from raw examples to training data
- A decision framework
- When to train, when to prompt, and when to use a bigger model
Booking
For dates, format, delivery arrangements and price, contact Noam about this workshop.
