2 Getting started with fine-tuning: A few examples

After reading chapter 1, where you discovered the value of specialized models, you should be eager to move beyond theory and build your first one. We’ll start with a working example that demonstrates something general-purpose models simply can’t do reliably: precise credit risk assessment with transparent reasoning.

2.1 Create your first specialized model

To understand what fine-tuning an LLM means in practice, let’s start with an example that feels concrete and immediately relevant. Suppose you’re part of a financial services team tasked with one of banking’s most critical responsibilities: credit risk assessment.

2.1.1 The dataset

2.1.2 The fine-tuning process

2.2 Parameter-Efficient Fine-Tuning (PEFT)

2.3 The fine-tuning mental model

2.3.1 The four-step fine-tuning process

2.3.2 How does this model perform on specific tasks?

2.3.3 How does this model perform on domain knowledge?

2.4 Creating a fine-tuning dataset

2.4.1 Collecting, cleaning, and structuring data

2.4.2 Labeling the data for supervised fine-tuning

2.5 A practical example of task-specific fine-tuning

2.5.1 Selecting the foundation model

2.5.2 Preparing the training dataset

2.5.3 Training the model on our dataset

2.6 References

2.7 Summary