# 6.96.0

January 16, 2025

## **What's new** 💡

### **Data Studio**

* Questions — Added [hints to all question types](https://github.com/vulcan-ai/gitbook/blob/datasaur/compatibility-and-updates/nlp-projects/lets-get-labeling/label-sets/README.md#question-hint) to provide clearer guidance.
* Row Labeling — Adjusted the colors for the row conflict indicator and made the label legend dynamically appear to improve visual clarity and usability.
* Span and Conversational Labeling Editor — Pressing ESC now closes the label dropdown when open, and clears span and label selection when the dropdown is not shown.
* Span Labeling — Adjusted the colors for the label set and special labels, and made the label legend dynamically appear to enhance visual clarity and usability.

### **LLM Labs**

* Deployment System — Completely redesigned the deployment workflow and interface.
* Direct Access LLMs — Added DeepSeek to our direct access LLMs.
* Evaluation Tools — Updated the manual evaluation export format to align with OpenAI's Direct Performance Optimization standards.
* Fine-tuning Features — Enabled fine-tuning capabilities to free trial users.
* Knowledge Base — Added new functionality allowing users to overwrite existing files in the knowledge base.
* Notification System — Enhanced the snackbar notification layout with better text wrapping and styling to eliminate horizontal scrolling.

## **Bug fixes** 🐞

### **Data Studio**

* Fixed an issue in Span Labeling where clicking "Go to Line" in advanced search results redirected users to a different file, but the line was not highlighted.
* Fixed an issue in Span Labeling where enabling Rapid Labeling Feedback caused labels to not appear immediately after applying a label following a line insertion.
* Fixed an issue in Span Labeling where selecting a different span after clicking a search result caused the view to jump to the search result labels.

### **LLM Labs**

* Fixed an issue where the deployed application ignored the advanced hyperparameters.
* Fixed an issue where the knowledge base encountered errors when uploading a thousand files.
