mirror of
https://github.com/SillyTavern/SillyTavern.git
synced 2025-06-05 21:59:27 +02:00
105 lines
5.3 KiB
HTML
105 lines
5.3 KiB
HTML
<html>
|
|
|
|
<head>
|
|
<title>TavernAI - Note - KobolAI Settings</title>
|
|
<link rel="stylesheet" href="/css/notes.css">
|
|
<meta charset="utf-8">
|
|
<meta name="viewport" content="width=device-width, initial-scale=1">
|
|
<link rel="preconnect" href="https://fonts.googleapis.com">
|
|
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin="">
|
|
<link
|
|
href="https://fonts.googleapis.com/css2?family=Noto+Sans:ital,wght@0,100;0,200;0,300;0,400;0,500;0,600;0,700;0,800;0,900;1,100;1,200;1,300;1,400;1,500;1,600;1,700;1,800;1,900&display=swap"
|
|
rel="stylesheet">
|
|
</head>
|
|
|
|
<body>
|
|
<div id="main">
|
|
<div id="content">
|
|
<h2>KoboldAI Settings</h2>
|
|
<p>Standard KoboldAI settings files are used here. To add your own settings, simply add the file .settings
|
|
in TavernAI\public\KoboldAI Settings
|
|
</p>
|
|
<h3>Temperature</h3>
|
|
<p>Value from 0.1 to 2.0. Lower value - the answers are more logical, but less creative. Higher value - the
|
|
answers are more creative, but less logical.</p>
|
|
<h3>Repetition penalty</h3>
|
|
<p>Repetition penalty is responsible for the penalty of repeated words. If the character is fixated on
|
|
something or repeats the same phrase, then increasing this parameter will fix it. It is not recommended
|
|
to increase this parameter too much for the chat format, as it may break this format. The standard value
|
|
for chat is approximately 1.0 - 1.05</p>
|
|
<h3>Repetition penalty range</h3>
|
|
<p>The range of influence of Repetition penalty in tokens.</p>
|
|
<h3>Amount generation</h3>
|
|
<p>The maximum amount of tokens that the AI will generate to respond. One word is approximately 3-4 tokens.
|
|
The larger the parameter value, the longer the generation time takes.</p>
|
|
<h3>Context size</h3>
|
|
<p>How much will the AI remember. Context size also affects the speed of generation.<br><br>
|
|
<u>Important</u>: The setting of Context Size in TavernAI GUI overrides the setting for KoboldAI GUI
|
|
</p>
|
|
|
|
<h2>Advanced Settings</h2>
|
|
<p>
|
|
The settings provided in this section offer a more detailed level of control over the text generation
|
|
process. It is important to be careful when making changes to these settings without proper
|
|
consideration, as doing so may result in degraded quality of responses.
|
|
</p>
|
|
<h3>Single-line mode</h3>
|
|
<p>
|
|
In single-line mode the AI generates only one line per request. This allows for quicker generation of
|
|
shorter prompts, but it does not produce responses that consist of more than one line.
|
|
</p>
|
|
<h3>Top P Sampling</h3>
|
|
<p>
|
|
This setting controls how much of the text generated is based on the most likely options.
|
|
Only words with the highest probabilities, together summing up to P, are considered. A word is then
|
|
chosen at random, with a higher chance of selecting words with higher probabilities.
|
|
</p>
|
|
<p>
|
|
Set value to 1 to disable its effect.
|
|
</p>
|
|
<h3>Top K Sampling</h3>
|
|
<p>
|
|
This setting limits the number of words to choose from to the top K most likely options. Can be used
|
|
together with Top P sampling.
|
|
</p>
|
|
<p>
|
|
Set value to 0 to disable its effect.
|
|
</p>
|
|
<h3>Top A Sampling</h3>
|
|
<p>
|
|
This setting allows for a more flexible version of sampling, where the number of words chosen from
|
|
the most likely options is automatically determined based on the likelihood distribution of the options,
|
|
but instead of choosing the top P or K words, it chooses all words with probabilities above a certain
|
|
threshold.
|
|
</p>
|
|
<p>
|
|
Set value to 0 to disable its effect.
|
|
</p>
|
|
<h3>Typical Sampling</h3>
|
|
<p>
|
|
This setting selects words randomly from the list of possible words, with each word having an equal
|
|
chance of being selected. This method can produce text that is more diverse but may also be less
|
|
coherent.
|
|
</p>
|
|
<p>
|
|
Set value to 1 to disable its effect.
|
|
</p>
|
|
<h3>Tail Free Sampling</h3>
|
|
<p>
|
|
This setting removes the least probable words from consideration during text generation, which can
|
|
improve the quality and coherence of the generated text.
|
|
</p>
|
|
<p>
|
|
Set value to 1 to disable its effect.
|
|
</p>
|
|
<h3>Repetition Penalty Slope</h3>
|
|
<p>
|
|
If both this and Repetition Penalty Range are above 0, then repetition penalty will have more effect
|
|
closer to the end of the prompt. The higher the value, the stronger the effect.
|
|
</p>
|
|
<p>Set value to 1 for linear interpolation or 0 to disable interpolation.</p>
|
|
</div>
|
|
</div>
|
|
</body>
|
|
|
|
</html> |