Training Data and Labels

Imagine an email service that must decide, for every incoming message, whether it is spam or not spam. The service has never seen the message that just arrived, so it cannot look up an answer in a table of known messages. What it needs is a system that has learned from earlier examples what spam tends to look like, and can then apply that learning to a message it has not encountered before.

In the spam case, the training dataset is a collection of emails. Each email comes with a label of either 'spam' or 'not spam' that a human has identified. The examples are the inputs the model will look at, and the labels are the answers it is supposed to learn to produce.

The two example emails above are invented for illustration, but the structure matches the source: each email text is the input, and each label is the correct answer stored alongside it. During training, the learning algorithm sees both halves of every pair. Later, during inference, the message the service is judging arrives without its label, so the label is the thing the model must produce.

The supervised learning workflow for predictive modeling is a sequence of stages that run in a fixed order. Stage one is training, in which a learning algorithm uses the labeled dataset to produce a trained model. Stage two is inference, in which the trained model is applied to new observations whose labels are unknown. Between these two stages sits a model evaluation step, which checks that the trained model meets the performance criteria we care about before it is allowed to make predictions in a real application.

Read the diagram from left to right. The labeled dataset and the learning algorithm together make up the training stage, and their result is a trained model. The model is then evaluated. Only after that does the workflow move to inference, where the trained model receives new observations and returns predicted labels. The trained model is the hinge of the whole process: it is the thing training produces and the thing inference uses.

A trained model is not automatically ready for use. Training produces a model that has fit the labeled examples, but the workflow requires an additional check before the model is applied to real observations. The source states that model evaluation should be added between the training and inference stages to ensure that the model satisfies the performance criteria before it is used in a real-world application.

The clearest way to see the workflow is to follow one concrete task from start to finish. Here the task is spam classification, and we track what data enters each stage and what comes out.

The two stages use data in opposite ways. During training, both the examples and their labels are known, and the learning algorithm uses that complete pairing to build the model. During inference, the model receives only the new observations. Their labels are unknown, and the model's job is to predict them.

The same workflow describes how Large Language Models are trained and used. The general pattern of training, evaluation, and inference stays the same. What changes is the behavior of the inference stage, which differs between the two kinds of LLM task the source distinguishes: text classification and text generation.