Frequency penalty lowers the probability that the model repeats the same words or phrases, proportional to how often they have already appeared. This is useful for preventing monotonous lists or loops. Closely related is the presence penalty, which penalizes a word's first occurrence regardless of frequency. Both parameters are usually set in a range from 0 to 2.
Frequency Penalty
Frequency penalty is a parameter that down-weights words that have already appeared often in the output to avoid repetition. Higher values lead to more varied text.