Configuring a Prompt in Experiments

Overview

The Experiments module enables efficient testing of multiple scenarios across different prompts, models, and evaluators simultaneously.

In this page we cover how Prompts can be configured within an Experiment context.

📘

See Setting up an Experiment to prepare your Experiment before model configuration.

Configuring your Prompt Template

There are two options to define your Prompt Template:

• Use the Messages column from your Dataset by enabling the Use messages as prompt template toggle (highlighted in red).
• Define Messages manually by using the message blocks in the Prompt column (highlighted in purple).

Import Prompt Template

Prompt Templates can be created separately and imported within an Experiment's Prompt Configuration.

To do so, make sure the chosen Prompt Template is available within your current Project and select the Import Prompt button.

You have the option to select a specific version of the Prompt.

The following modal opens:

Model configuration

You can select or change the model by clicking on the model pill.

Model Parameters are available to configure the model behavior.

📘

Models available in the Prompt configuration are taken from the Model Garden

Function Calling using Tools

Function Calling is available within your Prompt in the context of your Experiment

To add a Tool to your Prompt, use the Add Tool button. You can also use an existing Tool within your project using the Import Tool button.

📘

To learn more about Tools, see Tools in Experiments.

Knowledge Bases

Your Prompt can refer to a Knowledge Base.

Knowledge Bases need to be setup before, to learn more see Setting up a Knowledge Base.

📘

To learn how to use Knowledge Bases, see Using a Knowledge Base.