Model Steps
Last updated
Last updated
Model steps allow you to abstract one or more Generative AI model. You can match different size and cost models to your different use cases.
Select New Model Setp.
Step Name - Enter a name for this step.
AI Model Service - Select the model service this is running the generative provider.
AI Model Provider - Select the provider you want to use.
This will only show generative providers.
AI Model Credential ID - (Optional) Select the credential to use to authenticate against the model provider.
Max Characters to Send - Enter the maximum number of characters that can be sent to the model at once.
As a guide, 4 characters is a approximately 1 token for most models.
For public facing installations this should be the maximum size of the context you want to send to a model.
If this is set to 0, it will not be enforced.
Tools - Select the tools that will be sent to the model as part of this step.
Model - Select the model to use.
Max New Tokens - Enter the maximum number of tokens to generate.
For some cloud models the number of tokens you request will count towards your throttled limit not the number generated.
Temperature - Enter a decimal value between 0 and 1 to determine how the model will respond.
Temperature is a parameter that influences the language model's output. It determines whether the output is more creative or predictable.
A value closer to 0 will produce more predictable and less creative outputs.
Streaming - Choose if the response is returned to the user as it is generated (true) or once it is finished (false).
For user interactive chatbots this should be true.
If you are using the API for background agent tasks then you can set this to false.
Last Prompt Template - Enter a template that can be used for the last prompt sent to the model.
This can help re-anchor model responses. For example:
"{prompt} always respond using the provided results, and use the correct tool calling parameters." The users prompt is placed in the {prompt} placeholder.
