10.11.2023

New Releases, Enhancements, Changes + Arize in the News!

What's New

Prompt Playground

Uncover poorly performing prompt templates, iterate on them in real time and verify improved LLM outputs before deployment. Choose to edit the prompt variables, LLM parameters, or the prompt template itself. Then re-run the prompt to easily compare responses for analysis prior to implementing a new or revised prompt template. Learn more here.

Enhancements

Support for gpt-3.5-instruct-turbo

Users can now select to use gpt-3.5-instruct-turbo as their LLM of choice when using the Prompt Playground, or summarizing embeddings clusters.

Interactive Calibration Chart and New Precision Recall (PR) Curve

Users can interact with the calibration chart in the "More Charts" view of Performance Tracing. The chart now includes traffic bars and users have the ability to zoom into certain regions of their data, allowing for more granular insight. In addition, users can now select to view a PR curve for further analysis.

Y-Axis Controls On Dashboards

Users can now customize their Y-axis domain, customizing the min and max per-widget. Simply go into the edit form of a line or bar chart widget and hit the Display tab to control. Alternatively, users can employ auto and the charting library will compute what it believes to be the most comprehensible domain automatically.

Simultaneous UMAP visualization for multiple embedding features

Visualize your UPMAP for multiple embedding features for a multi-input model simultaneously, without needing to increase the number of predictions sent. To view multiple embedding features, select them in the "Dimensions" dropdown.

"Average" Option Added to Quantiles Plots

For additional views, users can now plot the mean, in addition to the previously offered median in quantile plots.

Python SDK v7.5.1

  • Core SDK is compatible with Python 3.6+ (previously 3.8+)
  • SDK now works with any version of pyarrow
Learn about Python SDK fixes and improvements here.

In the News

🎓 New LLM Course Updates

📚 New Paper Readings

Catch up on the latest in AI research papers with these new community readings:

Anyscale + Arize

We're proud to be a launch partner of Anyscale Endpoints! Through an early integration, Arize and Anyscale Endpoints enable developers to achieve LLM observability across different use cases on any cloud as their AI applications evolve.