Releases: awslabs/LISA
Releases · awslabs/LISA
v4.4.1
Bug Fixes
- Updated OpenSearch vector store creation to support private VPCs
System Improvements
- LISA now supports P5 Instances!
Acknowledgements
Full Changelog: v4.4.0...v4.4.1
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.4.0
b647273
This commit was signed with the committer’s verified signature.
estohlmann
Evan Stohlmann
Key Features
Image Generation
LISA now supports Image Generation capabilities!
Administrative Features
- Administrators can now configure and deploy models with the new IMAGEGEN classification type
User Experience Enhancements
- Users can customize image generation parameters including:
- Output quantity: Specify the number of images to generate per prompt
- Quality settings: Select between Standard and High Definition (HD) resolution
- Aspect ratio options: Choose from Square, Portrait, or Landscape formats
Image Management Functionality
- Comprehensive image handling options:
- Preview generated images directly in the interface
- Download individual images to local storage
- Copy images directly to clipboard for immediate use
- Perform bulk downloads of all images from a session to a zip file
- Regenerate variations using identical parameters
Persistent Storage Solution
- All generated images are automatically preserved in session-specific S3 storage
- Seamless retrieval of previously generated images when returning to a session
Directive Prompt Templates
LISA’s prompt library now supports directive prompt templates
User Template Management
- Users can now create and implement specialized directive prompt templates, complementing the existing persona prompt template functionality
- Seamless import capabilities allow for integration of directive templates into both active and newly created sessions
Flexible Access Control Options
Extended the existing permission settings enable users to designate directive templates as:
- Private resources for individual use
- Global assets accessible organization-wide
- Restricted resources with access limited to specific IDP groups
Workflow Integration
- Directive templates enhance structured interactions and standardized workflows across the platform
Acknowledgements
Full Changelog: v4.3.0...v4.4.0
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.3.0
v4.3.0
Key Features
RAG Ingestion Backend Overhaul
- Ingestion Job Tracking: Introduced a new DynamoDB table for tracking ingestion jobs using UUIDs. This enables real-time status queries and establishes a foundation for future monitoring and analytics.
- Execution Migration to AWS Batch: RAG ingestion workflows now run on AWS Batch with Fargate, removing the 15-minute timeout limitation of Lambda and enabling the reliable execution of large or complex ingestion tasks. This change also unlocks support for event-driven monitoring and job orchestration.
Benefits:
- Improved scalability and reliability of ingestion processes.
- Lays the groundwork for future enhancements such as parallelized embeddings and multi-step ingestion workflows.
Acknowledgements
Full Changelog: v4.2.0...v4.3.0
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.2.0
Key Features
RAG Updates
- The RAG stack has been updated to allow deployment without the necessity of deploying the UI.
- RAG ingestion and similarity search functionalities now support user tokens in addition to bearer tokens.
- Note: Administrators must deploy a Model and Vector Store via API using an Admin token before utilizing ingestion or search functionalities.
- Ingestion Pipelines can now be configured with a
0chunk size, enabling users to upload entire documents into a vector store as a single 'chunk'.
System Improvements
- Started building out our Python Unit Testing framework to enhance system reliability and performance.
Acknowledgements
Full Changelog: v4.1.1...v4.2.0
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.1.1
ffea676
This commit was signed with the committer’s verified signature.
estohlmann
Evan Stohlmann
Bug Fixes
- Upgraded LiteLLM so that SagemakerEndpoint hosted models will be supported again
User Interface Improvements
- Updated sessions UI to be more condensed and match the rest of the UI theme
- Save session configuration in DDB so when users re-opens session their settings persist
Acknowledgements
Full Changelog: https://github.com/awslabs/LISA/compare/v4.1.0..v4.1.1
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.1.0
Key Features
Image Processing
- LISA now supports LLMs that offer image analysis/input! During the model creation process, Administrators designate if a model is compatible with image input.
- Users are now able to incorporate images into their session context for supported image files. They can ask the compatible LLM questions about their images.
- LISA's message interactions have been restructured from langchain to system-managed objects, enabling the support of advanced message types. This breaks down messages into unique multi-part elements instead of one large text based message that is sent to the model.
Prompt Management
- Users can now create and modify personas for ongoing use across sessions. The visibility of these personas can be defined in the following ways:
- Personal use
- Specific Identity Provider (IDP) groups
- Public access (visible to all LISA users)
- When a user initiates a session, they can import and update the personas they have access to.
- Administrators can enable this functionality through the system configuration page.
User Interface Improvements
- When a request includes RAG documents, citations are now displayed inline in the ChatUI. They were previously only visible in metadata.
- A new landing page now has 'Quick Actions' that display if a user has yet to initiate a conversation.
- Users can now download their session history as a JSON file.
- The top menu options have been consolidated to minimize clutter.
System Improvements
- Upgrading third-party dependencies to leverage the latest features from our dependencies.
Upcoming Features
- We will be enhancing our Prompt Library to store user-defined prompt inputs in addition to persona definitions.
Acknowledgements
Full Changelog: v4.0.3...v4.1.0
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.0.3
Bug Fixes
- Resolved issue with subnets imports
- Resolved issue with custom model deployment
Acknowledgements
Full Changelog: https://github.com/awslabs/LISA/compare/v4.0.2..v4.0.3
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.0.2
Enhancements
- Revised base configuration to eliminate default RagRepository declaration. Important: Ensure config-custom.yaml contains an empty array declaration if no configurations are defined.
- Implemented multi-instance LISA deployment support within single AWS accounts. Customers may now deploy more than one LISA environment into a single account.
- Optimized data schema architecture to eliminate redundant reference patterns
User Interface Improvements
- Enhanced proxy configuration to support HTTP status code propagation for improved error handling
- Introduced configurable markdown viewer toggle for non-standard model outputs
- Implemented redesigned administrative configuration interface
- Enhanced session management:
- Removed UUID exposure from breadcrumb navigation
- Transitioned to last-modified timestamp display from access time
- Improved session loading indicators for enhanced user feedback
- Integrated document library refresh functionality
- Resolved critical Redux store corruption issue affecting state management overrides, reducing noticeable latency when fetching data in the UI
Acknowledgements
Full Changelog: https://github.com/awslabs/LISA/compare/v4.0.1..v4.0.2
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.0.1
244307a
This commit was signed with the committer’s verified signature.
estohlmann
Evan Stohlmann
Bug Fixes
Vector Store Management
- Enhanced UI to display default repository name when not specified
- Improved UI to show "GLOBAL" when no groups are assigned
- Refined repository schema regex to ensure valid input fields
- Optimized admin routing for RAG repository access
- Updated RAG Configuration table to align with config destruction property
- Resolved issue preventing creation of OpenSearch vector stores
User Interface
- Implemented consistent positioning of chat input at the bottom of the screen
Acknowledgements
Full Changelog: https://github.com/awslabs/LISA/compare/v4.0.0..v4.0.1
Contributors
dustins, estohlmann, and bedanley
Assets 2
v4.0.0
Our 4.0 launch brings enhanced RAG repository management features to LISA’s chatbot user interface (UI). Our new RAG document library allows users to view and manage RAG repository files. Administrators are now able to manage and configure vector stores (also known as RAG repositories), and document ingestion pipelines directly in the Configuration page without having to redeploy LISA.
Enhancements
RAG Repository Management
- Admins can create, edit, delete RAG repositories via LISA’s Configuration UI. Admins can also manage access through the UI. LISA re-deployments are no longer required.
- Admins can create, edit, delete new document ingestion pipelines via LISA’s Configuration UI. LISA re-deployments are no longer required.
- We added a RAG deletion pipeline that automatically removes S3 documents when deleted from RAG repositories.
- We introduced new API endpoints for dynamic management of vector stores and ingestion pipelines.
- Customers who previously configured LISA with RAG repositories (v3.5 and before) will be able to view these legacy RAG repositories in the Configuration UI. However, they will not be able to make any changes through the UI. Admins must continue to manage RAG repositories through the config file. We recommend that when you are ready, you delete any legacy RAG repositories through the UI. Then you will need to redeploy CDK which will automatically tear down the legacy repository’s resources. Then you will be able to recreate RAG repositories through the UI and re-load documents.
Document Library
- Added a RAG Document Library page in the chatbot UI. Users can download previously uploaded documents from the RAG repositories that they have access to.
- Users can also delete files from RAG repositories that they originally uploaded in the Document Library. Admins can delete any files through the Document Library. Files are also automatically removed from S3.
Note: As of LISA 4.0, new RAG repositories and document ingestion pipelines can no longer be configured at deployment via YAML.
Security
- Updated third-party dependencies.
Acknowledgements
Contributors
dustins and estohlmann