I understand that frequency penalty penalizes the repetition of tokens that have been used multiple times, while presence penalty penalizes tokens regardless of how many times they have been used.
However, I’m unsure about the meaning of the value range for these penalties. For example, in the case of Top P, a value of 0.7 indicates that only 70% of the weighted options will be considered by the model as possible outputs.
But when it comes to Frequency and Presence Penalty, it’s not as clear. What does a value like 1.5 represent? Thank You!