05 · Tokens, Context & Inference · Topic 05

What is Inference?

Inference is the process of running a trained model on an input to produce an output.

What is it?

Inference happens when a trained model is used to make a prediction or generate a result.

Training changes model parameters. Inference uses learned parameters on new input.

Training is studying for an exam; inference is using what you learned to answer a new question.

1 · Input Receive new data A prompt, image, audio clip, or other input.

2 · Represent Convert the input Make it readable by the model.

3 · Compute Run the model Apply learned parameters.

4 · Predict Produce a result Calculate a prediction or likely next token.

5 · Return Generate output Return a label, transcription, or response.

An image becomes a category such as “dog.”

A prompt produces a response through repeated next-token predictions.

An audio recording becomes a transcription.

Inference uses learned parameters; training changes them.

Inference is the broader execution process; next-token prediction can repeat within it.

Inference means model execution; reasoning describes possible problem-solving behavior.

Inference is what happens when a trained model is run on new input.

Independent explainer

not yet available

An Inference explainer video is not available yet.