What does “Repetition Penalty” mean?

A decoding setting that reduces the probability of tokens proportionally to how often they've appeared so far.

A decoding setting that reduces the probability of tokens that have appeared at all, encouraging new topics.

A decoding strategy that samples only from the k most likely next tokens.

A generation strategy that constrains output tokens to valid structures or policy-compliant choices.

An inference acceleration method where a small draft model proposes tokens that a larger model verifies in parallel.

An attention pattern where each token attends only to a fixed-size window of nearby tokens to reduce compute.