Skip to main content

Snorkel AI Data Development Platform v25.10 (LTS) release notes

Breaking changes​

  • Existing annotations are now tied to a specific task or batch. Any annotations on datapoints that are not currently in a batch or task will be deleted during the upgrade.

Deprecations​

There are no deprecations in this release.

Features and improvements​

Annotation​

  • You can now expand and collapse all fields in the annotation view to quickly locate specific data fields.
  • You can now view, collapse, and copy JSON fields directly in the annotation viewer, making it easier to work with structured data alongside text during annotation tasks.

Evaluation​

  • When you open an existing benchmark or create a new one, you now land on the benchmark overview page. From this page, you can access tabs to view information on runs and linked criteria.

User interface​

  • Collection pages and views now default to table view for a consistent user experience across the platform.

SDK​

  • You can now install the Snorkel SDK directly from PyPI using pip.
  • The SDK now supports annotation task functionality. See the SDK documentation for details.
  • The Dataset class now includes ground truth import and export methods.

Bug fixes​

Annotation​

  • Dataset columns beginning with "tokenized_" are now hidden in the annotation interface.
  • Fixed a typo in the label schema table.

Data management​

  • The SNORKEL_DATA datasource now uses the SNORKEL_INTERNAL_STORAGE connector.
  • Fixed an issue where uploaded folders briefly appeared and then disappeared when refreshing the page.

Known issues​

Data management​

  • Failures of ground truth import jobs are not surfaced in the UI.