Starter guide
How to Fine-Tune an LLM
A fine-tuning project begins with examples that show the behavior we want. Suppose the task is to turn meeting notes into action items. Each example contains a set of meeting notes and the action list that a good model should produce.
Some examples are used to teach the model. A different group is saved for checking its work afterward. Keeping these groups separate helps us see whether the model learned a useful pattern instead of remembering the examples it was shown.
What happens during fine-tuning?
Before training, we give the original model the saved meeting notes and keep its answers. This gives us a clear picture of how well the model already handles the task.
The model then learns from the training examples. When training is complete, we give it the same saved meeting notes again. We compare both sets of answers and check whether the new ones contain the right actions, leave out fewer details, and follow the expected format more often.