Fine-tuning
Level 2
Additional training that adapts a general model to a specific job or style.
Term 7 of 8 in Language models
In plain language
Instead of training a model from scratch, you take a finished one and train it a little further on your own examples. It comes out better at your particular task or tone.
Think of it like this
Hiring an experienced writer and then spending a week teaching them your house style.
Why it matters
Fine-tuning is one of two ways to make a model feel like yours. The other is giving it your documents at the time of asking.
Next terms in Language models
- Large language modelA model trained on enormous amounts of text to predict what words come next.
- TokenA small chunk of text that a model reads and writes, usually part of a word.
- PromptThe instruction and information you give a model.
- Context windowThe maximum amount of text a model can consider at one time.