ModelRefs / Repetition Penalty — AI Glossary
Repetition Penalty — AI Glossary
A decoding modifier that reduces the probability of tokens already present in the context, discouraging redundant output. 3).
Overview
Repetition penalty divides logits of previously generated tokens by a factor >1 (e.g. 1.1–1.3). Frequency penalty (OpenAI) penalizes in proportion to count; presence penalty applies a fixed penalty for any occurrence. Reduces the degenerate repetition loops common in greedy decoding but can degrade coherence if set too high.
Reference details
| Topic | inference |
|---|---|
| Also known as | frequency penalty, presence penalty |
| Last reviewed | 2026-06-24 |
Related terms
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Repetition Penalty — AI Glossary.
Frequently asked questions
What is Repetition Penalty?
A decoding modifier that reduces the probability of tokens already present in the context, discouraging redundant output.
Is Repetition Penalty the same as frequency penalty?
Yes — frequency penalty, presence penalty are common aliases for Repetition Penalty.
What concepts are related to Repetition Penalty?
Closely related concepts include sampling, greedy decoding, temperature.