Sillytavern frequency penalty will be penalized the most. Dynamic Temperature Min and Max temps, free to change as desired. You can vary temperature up to 1. It scales the token scores down, effectively penalizing words that appear too often, thus ensuring a more varied language throughout the dialogue. How many tokens from the last generated token will be considered for the repetition penalty. Frequency penalty makes it avoid common words and phrases, so it will speak in a more peculiar or sophisticated way. I recommend just using Hermes 3 405b free through openrouter (if/when it stops going down all the time). This can break responses if set too high, as common words like "the, a, and," etc. Frequency penalty just reduces the probability of a token. Frequency penalty adjusts the token's chance of reappearing based on its frequency of use in the text. However, the repetition penalty will reduce the probability because it's appeared too many times already. But too much may make it ignore the character description. 2 if you want to get more creative but weird messages and you can up Frequency Penalty if you feel its repeating itself too much as well, but your current settings don't look wrong to me. If the model repeats itself within one message, you can try increasing "Presence Penalty" or "Frequency Penalty". # How can I steer the story? Frequency Penalty: Decreases the likelihood of repeated words, promoting a wider variety of terms(i think). . Set the value to 0 to disable its effect. If the model repeats what's in the context, you can try increasing "Repetition Penalty" in the Completion Settings or you can try rephrasing the part of the context that's getting repeated. How are you feeling this morning? Yeah, good" After this good, we might get morning again as the most probable one. Presence Penalty Increases word variety. Imagine you have: "Hey man, good morning. Exponent, do not set Exponent higher than the default of 1. You can vary temperature up to 1. Presence penalty is so it repeats itself less. Presence Penalty : Decreases the likelihood of words that have already appeared in the text(i think again). Much smarter and better at RP in my experience plus you can tweak presence and frequency penalty. zpfbm qsggdh egu vfpbx nybj oqc xbgjlwi blr sjciy qreq