use case · wild
Normally the sampler chooses for you. Here you are the sampler. Start with a seed, and at each step GPT-2 offers its top candidates with probabilities — pick one and it becomes part of the context that shapes the next set. It's a hands-on tour of the possibility tree every LLM walks.
your text so far
GPT-2's next-token candidates — pick one
Notice how one choice narrows or opens the next: commit to "and suddenly" and the candidates lean dramatic; pick "and the" and they turn concrete. That branching is the whole game of text generation.
← Practical · Back to GPT-2 · Next: Multi-model →