Releases: awslabs/mlspace
Releases · awslabs/mlspace
v1.6.3
Enhancements
- Enhanced frontend end-to-end (E2E) test coverage for improved quality assurance
- Upgraded support for several dependency libraries, including:
- CloudScape
- Webpack
- Micromatch
- Express
- Expanded documentation on security policies and roles, providing System Administrators with clearer understanding of MLSpace user capabilities
Bug Fixes
- Resolved an issue related to unsafe calling of an unsupported IAM feature in certain regions
Acknowledgements
Full Changelog: v1.6.2...v1.6.3
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.6.2
Release 1.6.2
Key Features
Group Membership Auditing
- In the group details page users will now be able to see when users are added and removed from a group
- This information will contain who was added, when they were added, who conducted the action, and when the action happened
- The data will persist in Dynamo even if the group is deleted so if an Admin has to do user access research the data will still be available
Enhancements
- Remove deprecated constants from CDK
- Removed un-used CO permission scheme from all Back-End code
- MLSpace will now display the number of group members in a group in the table display
- Updated the AppConfig confirmation screen to show user-friendly field names
Bug Fixes
- Fixed bug where AppConfig would potentially error out when getting pulled into the UI
- Merged in a CVE fix by a 3rd Party
Axios
Acknowledgements
Full Changelog: v1.6.1...v1.6.2
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.6.1
Key Features
Datasets
- Multiple Groups can now be assigned to a Dataset
- All of the Datasets that a Group has access to are listed on the Group’s details page
- Admins now have the ability to easily monitor Datasets access in MLSpace. Orphaned Datasets are flagged for Admins' attention to prevent loss of data and ownership.
Group/Project Association
- Groups can now be added to Projects as Project collaborators or owners. The associated role will be inherited by every member of the Group
- All of the Projects that a Group has access to are listed on the Group’s details page
Enhancements
- The latest release notes are now displayed on the login page as another example custom component
Bug Fixes
- System Banner text is now bold
- System version is now shown without scrolling if banner is enabled
- Group breadcrumbs weren't updating correctly
- Dynamic Role user polices now use correct s3 group resources
- Edited Batch translate S3 input selector to only allow directory input
Acknowledgements
Full Changelog: v1.6.0...v1.6.1
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.6.0
Key Features
MLSpace Groups
- Administrators can now create Groups in MLSpace!
- Administrators can manage user membership for the Groups.
- Users can now associate Groups with Datasets, expanding permissions beyond the Global, Project, and Private options. This allows teams working across projects to access Datasets easily.
- Users can get a detailed view of the Groups to see which Groups they are a part of and the other members of their Groups.
User Details
- Admins can now drill down to get more details about individual users
- This will allow admins to see what access a user has to projects and groups on a single page
Bug Fixes
- Standardize enum and EnvVariable usage
- Documentation readability improvements
- Various fixes across the dataset and project management pages
Coming Soon
- 1.6.1 is targeted for end of July where the team will be delivering the ability to add the newly released groups as project collaborators or project owners. This permission scheme will be inherited by every member of the group. The team will also be adding multi group support for datasets, where datasets can be shared with multiple groups for use within their member's projects.
Acknowledgements
Full Changelog: v1.5.3...v1.6.0
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.5.3
Bug Fixes
- Updating labeling job output path to be more accommodating for trailing slashes
- Add notebook policies to dynamic roles upon every deployment
Acknowledgements
Full Changelog: v1.5.2...v1.5.3
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.5.2
Key Features
- Added conditional policy for
VolumeKmsKeyIdonly for supported instances - Introduced a Lambda function to dynamically update the above policy with available instances that will execute daily so as new instances are added they will be accounted for
- Added some automated deployment enhancements to our GitHub actions that allow us to easily roll out releases
Bug Fixes
- Fixed a bug where terminated EMR clusters were disappearing from the list view unexpectedly
- Fixed a race condition that was encountered during EMR cluster creation when navigating to details page after a creation success response was received
- Fixed a bug where the notebook policy wasn't getting granted permissions for certain actions when
MANAGE_IAM_ROLESwas set toTRUE
Acknowledgements
Full Changelog: v1.5.1...v1.5.2
Contributors
dustins, estohlmann, and douglas1850
Assets 2
v1.5.1
Bug fixes
- Fixed bug where ARNs for the instance constraint policies weren't being set as environment variables correctly for lambda's that used them
- Fixed bug where the HPO Wizard wasn't correctly validating user input before sending the jobs to the BE
Full Changelog: v1.5.0...v1.5.1
v1.5.0
Key Features
In-App Configuration
MLSpace Administrators can easily manage commonly customized settings directly in the MLSpace user interface. This feature removes the need for code changes and redeployments.
Administrators can now:
- Manage available EC2 instance types for:
- SageMaker Notebooks
- Endpoints
- Transform Jobs
- Training Jobs
- Activate or deactivate access to the following services:
- Amazon Translate
- Amazon GroundTruth
- Amazon EMR
- Define EMR configurations by controlling available applications and defining cluster types
- Restrict project creation to Administrators for installations in less permissive environments
- Easily update and configure the application banner
- Save all historical admin configuration settings:
- View history in-app via the configuration history table
- Audit changes, including timestamps and authorship
- Roll back to any previous configuration
- Import/Export settings to manage multiple environments
Dataset Management Enhancements
- Preserve file structure during bulk uploads
- Enable drag-and-drop functionality for uploads
- Enhance upload notifications with a progress indicator
- Display a DatasetSelector warning when files don't exist
- Make the dataset table use serverFetch and proper loading actions
UI Cleanup
- Standardize input debouncing in the frontend
- Remove excess background refresh intervals
- Prevent breadcrumbs from reloading the entire app layout
- Add loading icons for background refreshing
- Add additional spacing between Notebook action buttons
- Provide more user-friendly error messages for quota issues or name conflicts
- Allow prefix selection for batch translate input
- Clean up labeling job creation validation
Documentation Updates
- Basic documentation for creating a GroundTruth workforce with Keycloak
- Minor grammar and wording cleanup
Full Changelog: v1.4.1...v1.5.0
MLSpace v1.4.1
Bug fixes
- [AIML-ADC-8888] Fixed bug where users were unable to select custom lifecycle config during notebook creation
- [AIML-ADC-8887] Fixed bug where terminated EMR clusters were being returned as selectable clusters in the notebook creation modal
MLSpace v1.4.0
Features/Improvements
- Dataset improvements
- Add support for the creation of empty datasets
- Training jobs started from notebooks will automatically create datasets if the output is written to an S3 path not currently associated with a dataset
- Added a warning when empty datasets are selected for input channels
- Added new dataset browser component and reworked all forms that allowed users to select input/output paths from datasets
- Added support for additional built-in training job algorithms
- Linear Learner
- Object Detection (MXNet)
- Semantic Segmentation (MXNet)
- XGBoost
- Added new config CLI to help users generate a config file with environment specific configuration
- Added support for deploying ARM based lambdas
- Add support for whitelabeling the application and documentation
- Switch from mkdocs to vitepress for documentation
- Added generic configurable system banner
- Added landing page instead of immediately redirecting to login
- Added optional support for OIDC Client Secret flow
- Added tuning job source/parent field to training job details view
- Added ability to clone training jobs
- Added ability to clone HPO jobs
- Added darkmode
- Added project resource counts overview to project details view
- Added ability for admins to generate reports based on specific projects or users
- Added version label to side navigation
- Batch and real-time translation forms will now remove the selected source language from the target language selection options
- Added the ability to select a specific subnet during model creation
- Added input data s3 location to traing job defition details view in HPO details view
- Increased default page size for all tables and began tracking table page size preferences in user preferences
- Removed format and classification field from datasets
- Added the ability to specify a subnet as part of EMR Cluster creation
- Replaced Quill with tiptap in labeling job creation workflow
Bug fixes
- Fixed bug where training job form could be submitted without client-side validation running
- Fixed bug where continuous range parameters were not properly validated client-side
- Fixed accessibilty issues with labeling job creation form
- rich text editor buttons lacking proper keyboard navigation support, - accessible button labels, accessible icon labels
- task timeout inputs missing accessible labels
- dynamic label control missing accessible labels
- Fixed bug where batch translate jobs run in us-iso-east-1 would fail due to unsuported
autodetection was selected - Fixed bug where creating an HPO job with multiple training job definitions results in duplicate metrics values
- Fixed bug where input data configuration "mode" for training/HPO jobs was required when it should have been optional
- Fixed an issue where users were unable to create training jobs from the UI without changing the default instance type
- Fixed bug where the delete button on global datasets was enabled for users who did not known the datasets. Deletion would fail but the button should not have been enabled
Full Changelog: https://github.com/awslabs/mlspace/commits/v1.4.0