6.104.0

March 13, 2025

What's new 💡

NLP

• Bottom-Level Label Selection in Span Labeling — Added an option to allow label selection only for bottom-level labels when applying labels.

• Span Labeling:

  • Automatically assign colors to labels when adding a new label without a predefined color.

  • Improved labeling performance for projects with a large number of labels.

• Advanced Search in Span Labeling — Improved search performance for faster results.

• Analytics Update for Pre-Labeled Data — Integrated pre-labeled data into analytics calculations and adjusted metrics for consistency across total labels applied, accepted, and rejected, affecting CRB and various analytics charts.

LLM Labs

• Usage — Refactored our embedding usage tracking; embedding usage will now be consistently tracked at the embedding service layer, ensuring both evaluators and vector store operations are included.

• UI Improvement — Adjusted the decimal numbering in LLM Labs.

• Knowledge Base — Changed our document chunker to HybridChunker to prevent context overflow.

• Direct Access LLMs — Integrated Anthropic as a new provider in Direct Access LLMs.

Bug fixes 🐞

NLP

• Fix an issue where replacing the question set did not reset all draft answers in a project.

LLM Labs

• Fixed an issue where the sorting mechanism on the deployed application page was not working correctly.

• Fixed an issue where the circular loading indicator appeared on unselected prompts.

• Fixed an issue where ranking evaluations were able to be used as a fine-tuning dataset.

• Fixed an issue where a blank page occurred when navigating to account settings.