MY UNNOFICIAL PARAMETER GUIDE: I generally use my presets based of my current understanding of the parameters and what kind of character I'm playing with. If the character has example dialogue in it, is an RPG, or involves complex instructions, the parameters may need to be lower or a bit higher overall. Parameter Summaries

• Max Tokens: Limits output length. If set too low, the output can be truncated.
• Temperature: Controls randomness. Higher values make output more random; too high can lead to incoherence.
• Top P: Limits to top cumulative probability tokens. Lower values reduce the number of possible responses, making output more focused.
• Top K: Limits to top K tokens. Higher values can reduce repetition but also quality and coherence.
• Repetition Penalty: Reduces repeated tokens. If set too high, it can cause inaccuracies or altered details.

Rule of Thumb

• Max Tokens: Keep at 200; you can easily adjust response wordiness and such via other parameters and system prompt.
• Temperature: Keep low - the randomness does not mean it will give better or more interesting outputs.
• Top K: Don't use on characters that need precision or have formatted text. Avoid with Nephra 8B and Pharaoh 24B
• Repetition Penalty: Keep at 1.15 or lower in the beginning of the chat. (first 5 messages) If repetition gets worse, increase to 1.2. If repetition becomes even worse, switching from much higher values to lower values throughout the chat can help. At 1.15 it is unlikely to alter information or important memories/facts of the character's personality, as well as real-world information such as gender and occupations of ECs that aren't in your character. Values below 1.15 may cause issues with adapting the scene to unexpected events and new formats.

Now, at the beginning of the chat for the first three to five messages, you may want to keep repetition penalty at a maximum of 1.1. Then, feel free to swtich it up to 1.15.