Frequency & Presence Penalty Settings
The Frequency Penalty in Settings, controls the diversity of generated text by penalizing new tokens based on their frequency in the text.
Settings can be between 0-2.
Positive values decrease the likelihood of repeating the same line by penalizing frequently used tokens.
The parameter modifies the probability distribution of the model to generate less common words, encouraging novelty.
The Presence Penalty is another parameter that discourages generating words already in the input text.
Both parameters increase diversity and encourage novelty in generated text.
Title Prompt Example
I want the title to be original and unique and just borrow from the variable data I give it.
These settings do not matter so much for small chunks of text like titles, but if you find that there are patterns you do not like in your titles, try tweaking these up a little.
If you have a cluster of similar keywords, you will notice more variety in the output.
First Draft Example
For the first draft, while I don’t want it to be too competitive, I also want it to develop a tone and be somewhat consistent, so I have opted for 1.2 for the presence penalty.
I also don’t want it making up facts and I do want it to borrow from whatever resources it has on the topic, so I have reduced the Frequency Penalty.
I must add, this is all experimental, so tweak these and see what kind of output you get. You will find that some topic clusters or sites will require different settings. Be flexible.