Prompt Studio

Goal

Prompt Studio helps transform datasets of any format into structured datasets where each row is a fine-tuning-ready example. This ensures compatibility with AI model training requirements while simplifying the process.

Key Features

• Structured Output: Automatically generate a dataset where every row is a valid training or fine-tuning example for AI models.
• Customizable Prompt Roles: Configure prompts for system, user, and assistant to handle diverse use cases.
• Dynamic Macro Embedding: Define and embed macros directly into prompts using dataset fields, NQL expressions, or literal values.
• Interactive Preview: Validate outputs row-by-row in a conversational format to ensure accuracy.

How to Use Prompt Studio

Step 1: Select a Dataset

1. Start by selecting a dataset from the Dataset module in the Prompt Builder.
2. This dataset will serve as the source for creating rows aligned with fine-tuning requirements.

Step 2: Define Prompts

For each participant (System, User, Assistant):

1. Click the Select button under the respective role.
2. Write the prompt text, incorporating placeholders for macros (e.g., {{macro_name}}) where dynamic values should be embedded.
3. After defining macros in prompts, configure them to pull the appropriate values:
   • Field: Map the macro to a column in the dataset.
   • NQL (Narrative Query Language): Define a query expression that dynamically generates the macro value.
   • Literal: Provide a fixed text value for the macro.
4. Save the configuration for each prompt.

Step 3: Preview Outputs

The Preview screen provides an interactive validation tool:

1. View the prompts for each row in the dataset in a conversational format, showing the system, user, and assistant messages.
2. Use the Prev and Next buttons to navigate through dataset rows and confirm all macros are correctly resolved and formatted.

Step 4: Save the Prompt Mapping

Once the prompts and macros are finalized:

1. Use Data Studio to generate a new materialized dataset where each row adheres to the structure defined in the prompts.
2. This structured dataset is ready to be used in the fine-tuning process for AI models.

Output Format

The resulting dataset will contain rows formatted as structured conversations, with prompts dynamically filled based on the macros and configurations. This ensures a seamless transition to fine-tuning AI models.