Using response preferences in General knowledge step
This guide covers how to configure response preferences to refine and optimize your outputs in Amazon Quick Flows, providing flexibility in response optimization based on your specific use case requirements.
Key highlights
- Simplified model selection
-
Flow builders get a benefit-based preference selection for their output refinement where they can choose from 2 modes - Faster responses or Versatility and performance. This reduces cognitive load for builders and creators can focus on their objectives rather than technical model comparisons.
- Intelligent model selection in runtime
-
Depending on your output preference, flows service will automatically select the most appropriate model based on real-time context size, task and multi-modal requirements.
- Modality supported for general knowledge step
-
Input: Text/document files, Image or Video, output: Text. Users can upload up to 50 MB of document files, 1GB of video files, and 4.5 MB of image files as inputs.
Getting started: Response preferences in flows
When building flows in Amazon Quick Flows, you can select response preferences to optimize performance for your specific use case. The response preference interface allows you to choose the most appropriate optimization based on your requirements for speed, versatility, and performance.
To select response preferences:
Navigate to your flow configuration
Add a General knowledge step
Access the response preference options
Choose from Faster responses or Versatility and Performance
Configure additional settings as needed
Configuring output types: Text vs Image
Different Amazon Bedrock models support various output formats. Configure your output type based on your application needs:
Text outputs
Text outputs are optimized for natural language generation and support both structured and unstructured text with variable length responses based on model capabilities.
Image outputs
Image outputs provide visual content generation capabilities with support for various image formats and resolutions, including integration with text prompts for image generation.
Advanced model settings: Creativity slider, Exclude, and Seed
Fine-tune model behavior using advanced configuration options:
Creativity slider
The creativity slider controls the randomness and creativity of model outputs. Lower values produce more deterministic results, while higher values increase variability and creative responses.
Exclude settings
Exclude settings allow you to specify content or patterns to exclude from image outputs, helping maintain content guidelines and restrictions with customizable filtering based on your requirements.
Seed configuration
Seed configuration enables reproducible outputs for testing and consistency. Use specific seed values to generate consistent results, which is useful for debugging and quality assurance workflows.
Multi-modality support using Amazon Bedrock models
Leverage models that support multiple input and output modalities:
Text-to-text: Traditional language model interactions
Text-to-image: Generate visual content from text descriptions
Image-to-text: Extract information or descriptions from images
Multi-modal combinations: Process both text and image inputs simultaneously
File uploads using general knowledge
Amazon Quick Flows supports various file types and processing capabilities with Amazon Bedrock models. Supported formats include documents, images, and structured data files with processing options to extract text, analyze content, or generate summaries. Integration workflows seamlessly incorporate file content into model prompts, though you should refer to model-specific file size restrictions.
Total context limit supported for Amazon Bedrock models
Understanding context limitations helps optimize your applications. Context window sizes vary by model type and version, so monitor input and output token usage. Use optimization strategies and techniques for working within context limits while balancing context size with response speed for performance considerations.
Note: If you don't see response preferences, contact admin
If the response preference options are not visible in your interface:
Verify your user permissions and access levels
Contact your system administrator to ensure "Enable bedrock model usage in General knowledge step for output refinement" is enabled
Ensure you're using the latest version of the Amazon Quick Flows interface
For additional support and configuration details, administrators can refer to the comprehensive capabilities documentation.
