Information valid from AMI version 8.0 SP24 (24.2)
Summary
The AI Configuration System Preference lets you set the AI generated values for attributes.
Gadget Action Menu Items
Connection details
You must define the location of the AI provider and an API Key for your account:
- URL
- AP Key
Parameters
The parameters are settings that affect how a value is generated. These may be different from one model of AI to another, but common ones are:
- temperature - a numeric measure of how creative a generated value should be, with high numbers requesting more creative, random values and low numbers requesting more predictable, consistent values.
- top_p - a numeric measure of how broad the choice of words should be, with high numbers allowing a wide choice of vocabulary and low numbers limiting the value to common or repeated words.
- max tokens - limit the number of tokens used to request and return a generated value. A token is a word fragment or piece of punctuation
Your Agility system may have different parameters available, or use different terms (labels) for them. You can set "usage" explanations for each parameter and provide minimum and maximum value ranges to guide users.
Add AI parameter
Adds a parameter, and a default value for it, to send for the generation requests.
Each parameter must have at minimum a Key and an associated Value but you may specify additional information to make it easier to use in Agility.
Each parameter must have a unique key.
- Key - the name of the parameter used by the AI Service.
- Value - a default value for the parameter. This value can be changed on the AI tab of the Attribute Editor to fine tune requests.
- Label - a user-friendly name for the parameter.
- Usage - a description of the parameter that your users can view by hovering over it.
- Min / Max - minimum and/or maximum values for numeric parameters. If a user tries to enter a value outside the range it will be rejected.
- Slider - if an parameter has both a Min and Max value set then you can choose for it to be displayed to users as a slider when fine tuning requests.