February 14th, 2025

Changelog (v2.170.0)

📢 Private Beta Release: LLM Data Labeling

We’re excited to announce the private beta release of our LLM data labeling feature! 🎉

This new capability enables you to create high-quality training data for model alignment through two labeling modes:

• LLM Dynamic Labeling – Generate conversations by prompting models, rank responses, and refine outputs.
• LLM Static Labeling – Import and label existing conversations, such as logs from production environments.

Interested in trying it out? This feature is in private beta—contact us for access.

For more details, check out our full guide: LLM Data Labeling Documentation.

Other new features

• [Issue deletion] Project admins can now cancel issues that are no longer valid.

In the SDK, this can be done programmatically using the update_issue_status method by setting the status to CANCELLED.

Change

• [SDK] The copy_project method has been optimized for better performance when copy_assets=True. Additionally, the copy_quality_settings and copy_json_interface parameters are now deprecated, as these settings will be automatically copied to maintain project consistency.

Fix

• [Cloud storage] We fixed an issue where the Queue page failed to load after removing a data connection in a project with already synchronized data. Now, the Queue page loads correctly. If an asset is missing its file, an error message will be displayed when opening it. You can still delete unusable assets.
• [Vertices on rotated image] Fixed an issue where annotation vertices were incorrectly altered when exporting labels from rotated images in the Kili format.
• [Filters] Fixed an issue where asset filters did not work correctly for decimal numbers in transcription items.