An LLM does not write a response all at once. It generates text incrementally, one token at a time.

At each step, the model receives the tokens generated so far, runs a forward pass through the transformer, and produces raw scores for every token in its vocabulary. Those scores are converted into a probability distribution. A decoding strategy then chooses the next token.

That loop is the mechanical core of text generation:

context -> model -> token scores -> decoding -> next token -> updated context

The details matter. Temperature, top-p, stop tokens, context length, and output limits all plug into this loop. When an LLM repeats itself, drifts off task, fabricates a citation, or stops mid-sentence, the cause is often visible in the generation mechanics.