Concept重复惩罚 / repetition penalty
Repetition Penalty
An LLM decoding parameter that suppresses the model's tendency to reuse the same phrases by penalizing the probabilities of tokens already generated, reducing redundancy and boilerplate in output.
Timeline (last 90 days)
Aug 25
High temperature combined with a missing repetition penalty can easily cause the model to fall into repetitive loops or generate meaningless content.
Unverified60%