01.04.2024
New Releases, Enhancements, + Changes
What's New
Granularity Selector
Analyze performance by hour, day, week or month using the new granularity selector. Options default to the optimal granularity for a given time range. The granularity selector is available on the monitors page, all the relevant model pages, and the dashboards page.
Slice Performance Over Time for Tracing
Users can now view the performance of individual feature slices over time. In the performance tracing tab, select a feature and click "Slice Performance Over Time". This will plot the performance of the top five slices of that feature. Correlate performance issues with individual slices, further debug your features, or better understand trends over time.
Google Gemini Support
Users can now utilize Google's new Gemini model via the Vertex AI integration in Arize's Prompt Playground.
Phoenix Demo Page
Check out LLM Tracing via Phoenix's UI via https://phoenix-demo.arize.com/tracing. For more info, head over to Phoenix docs on tracing your LLM application.
Enhancements
Generative LLM Overview Refresh
The generative LLM model overview page has undergone a couple of changes to increase ease of use:
Quick Actions cards linking to various in app workflows
Model Schema is now included on the left side navigation
Python SDK v7.8.0
This release enables latent actuals for generative LLM models. In addition, users will now be alerted when their data files are too large.
Python SDK v7.9.0
This release brings to life a new model type: multi-class. Users can now log not only the classes by their classification models, but also the non-selected ones, giving a detailed view of how the model operates. Currently, only the record-at-a-time logging option is supported, but stay tuned for the release of pandas batch ingestion.
The latest ebooks, self-guided course modules, and technical posts on topics like LLM evaluation and beyond:
LLM evaluations: everything you need to know
OpenAI Function Calling: Benchmarking, Explanations
Prompt Templates & Evaluating Prompts
Text To SQL with AI: A Study in Zero-shot, Single-domain, and Cross-domain Settings
Mistral AI: A Deep Dive Into the Latest Models
Last updated
