●Change8

arize-phoenix-v13.0.0

Breaking Changes
📦 arize-phoenixView on GitHub →
⚠ 1 breaking✨ 39 features🔧 1 symbols

Summary

Version 13.0.0 introduces extensive enhancements to LLM evaluation capabilities, including new built-in evaluators, improved playground features, and UI updates. This release contains breaking changes related to dataset evaluators.

⚠️ Breaking Changes

  • Dataset evaluators functionality has been significantly changed or removed. Review usage of dataset evaluators and update code accordingly.

Migration Steps

  1. Review and update any code relying on the previous implementation of dataset evaluators.

✨ New Features

  • Added autocomplete functionality to the LLM evaluation prompt editor.
  • Added available tools information to experiment output.
  • Added built-in LLM evaluator configurations to GraphQL.
  • Added support for custom providers in the model menu.
  • Added dataset deep link functionality after selection from the playground.
  • Added default output configuration for LLM evaluators.
  • Added descriptions to built-in components.
  • Added evaluation outputs to the playground interface.
  • Added evaluator count display to the tab.
  • Added evaluator kind token.
  • Added evaluator label to all evaluator prompts.
  • Added evaluator preview mutation.
  • Added EvaluatorKindToken.
  • Added evaluators table to the dataset evaluators page.
  • Added examples route with an examples table.
  • Added explanation toggle to the evaluator form.
  • Added input mapping support to built-in evaluators.
  • Added JSON parse toggle to the JSON distance built-in evaluator.
  • Added metadata to the evaluator database table.
  • Added model search capability to the model menu.
  • Added model information to evaluator tables.
  • Added more built-in evaluator forms and improved the flattening utility.
  • Added more built-in evaluators.
  • Added support for OpenAI API type (Chat Completions vs Responses API).
  • Added optional description field for new evaluator creation.
  • Added output configuration display to built-in evaluators.
  • Added pre-built LLM evaluators to the evaluator creation menu.
  • Added prompt information to the evaluations table.
  • Added SwitchableEvaluatorInput to enable customizable evaluation inputs.
  • Added the ability to create examples within a chain.
  • Added tool response handling evaluator template.
  • Added user ID tracking for evaluators.
  • Appended messages parameter in the playground.
  • Implemented Builtin Evaluator Config Overrides.
  • Implemented Builtin evaluator table view.
  • Cleaned up the preview UI to show a full annotation value.
  • Cleared table state when dataset or splits change.
  • Collected all JSON path segments when flattening example keys.
  • Introduced composite field for model + parameters.

Affected Symbols

dataset evaluators