Overview

Thanks for helping us test different setups for WyvernChat’s free queue models. By rating each output, you’ll help us decide which features to refine and which to keep. We’ll look at five key things:

1. Coherence: How well the text flows and makes sense.
2. Repetition: If the response repeats words or ideas too often.
3. Creativity: How original or interesting the response is.
4. Godmod/Agency Loss: How much the “bot” respects your persona’s freedom.
5. Character Adherence: How closely the output matches the character’s traits and story.

We have many system prompts, models, and sampling settings to try. Your scores and comments show us which setups work best. We do not use personal data to train our models. Our main goal is to keep WyvernChat easy to use and open to everyone.

Objectives

• Stay Consistent: Use the same scoring rules every time so we can compare different tests. This helps us see which setups are good and which need more work.
• Give Clear Feedback: We need your number scores, but any extra details or notes can help us understand the best and worst parts of each test.
• Keep It Private: We never ask for your email, and if you share screenshots, you can cover any personal details first. We store and review all data on local drives and in Notion, which are protected with secure passwords.
  • Weekly Check: SopakcoSauce checks responses once a week to make sure nothing personal is left in the files. If you forget to blur or remove something private, he’ll do it for you.
  • No Personal Data in Training: We do not use personal information to train any of these models. Our focus is on making WyvernChat easier to use for everyone.

Before You Begin

1. Know the Five Scoring Points
   • Coherence: Logical flow and clarity of the text.
   • Repetition: Avoidance of excessive or redundant phrasing.
   • Creativity: Is there a good amount of variety in each “reroll”?
   • Godmod/Agency Loss: Does it talk for you?
   • Character Adherence: Does it stick to each character’s background and personality?
2. Select the System Prompt and Model:
   • From the provided repository of system prompts, choose one to test. These prompts guide the model’s style.
   • Select an LLM model (like LooseCannon, EVA-Qwen, or StellarOdyssey) and a sampling preset (such as Gunpowder or AHDAM-Wren).
   • Note your choices on the scorecard so you know exactly what you tested.

Filling Out the Scorecard

Each scorecard covers one combination of Model, Prompt, and Sampling Parameters. Here’s what to do:

1. System Prompt + Model + Sampling Parameters

   • Identify which model you are testing.

   • Indicate which system prompt you selected.

     • You can see a list of options for by clicking “System Prompt”

       

   • List the sampling parameters used.