Files
SillyTavern/public/notes/4.html
2023-04-22 00:21:27 +02:00

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&amp;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>