Model Settings
These settings help control how the model generates the character’s responses.
Model — choose the appropriate model for the character’s operation.
Temperature — controls the creativity level of responses; recommended range is 0.5 to 1.1.
Top-P — limits word selection to the most probable set covering probability P. For example, at P=0.9, the model chooses words covering 90% probability.
Top-K — limits word selection to the K most likely words. For example, at K=50, the model chooses only from the top 50 words.
Total Context Tokens — the maximum context size the model considers when generating a response.
Max Response Tokens — maximum response length. On average, 100 tokens equal about 70 English words.
Presence Penalty — reduces the chance of repeating words already used, as if saying to the model: “You’ve already said this, don’t repeat.” The higher the value, the stronger the effect.
Frequency Penalty — penalizes frequent repetition of the same words, like saying: “You said this three times — stop.” The higher the value, the stronger the influence.
Last updated