Generate Q&A

Introduction: User can now generate Q&A from the trained content via topic prompt or file upload in the new Evaluate tab of AI Admin.

Key Benefits:

• Allows users to ensure that the generated answers align with the trained content
• Identify gaps or inconsistencies in the training data for corrective actions
• Modify answers to retrain the LLM with expected answers

Steps to Generate Q&A via Prompt

1. Enter file name for export
2. Select All or Individual Content tags.
3. Enter Topic Prompt & Question count for generation.
4. Select LLM Configuration (Model, Answer Type & Precision)
5. Click on Generate button
6. View generated source file on in the generate prompt tile.

Additional Info:

• Max 50 questions allowed in a single generate operation via prompt
• Default values visible in the LLM configuration will be same as configuration present in workspace settings tab
• Based on available content & context generated questions can be less than or equal to the question count entered by the user

Steps to Generate Q&A via Upload

1. Select All or Individual Content tags.
2. Upload questions in sample csv format.
3. Select LLM Configuration (Model, Answer Type & Precision)
4. Click on Generate button.
5. View generated source file on in the generate upload tile.

Additional Info:

• Max 100 unique question are allowed in csv for answer generation. Duplicate questions will be ignored.

Tile View In Evaluate Tab: For every Generate operation a new tile will be generated and source file will be present in the tile. Source file can be downloaded in csv format & if content tag is trained with new content user can regenerate the answers for same set of questions by clicking on regenerate icon.

Key Highlights:

• Content Sources Covered: Docs, Website URL, Text, Files. Product Catalog is not referred.
• Data Retention: 365 Days
• Workspace Availability: Both Generic & Commerce Workspaces
• Generate Q&A csv compatible with teach module
• Max 50 generation request allowed in a workspace. No limit on regeneration. To create new generation request, user can delete existing tiles when 50 limit is reached.
• Only 1 generation/regeneration request allowed at a given time in the workspace.
• User can change LLM configuration parameters (Model, Answer Type & Precision Threshold) to evaluate the impact on generated Q&A.