Skip to main content
Version: v0.16.0

Release notes

v0.16.0 | May 10, 2023​

At the core of the new version, H2O Model Validation improves and expands on its supported validation tests. These new improvements and expansions further expand the goal of offering a platform that enables you to assess the robustness and stability of your models and datasets. Among other updates, you can now create a Connection to My AI Engine Manager.

To learn more about the new release, observe the below subsections.

UI & UX

  • New: The H2O Model Validation landing page now has a new design that provides an array of informational points about the instance of the application. For example, on the landing page, you can now observe the number of Connections, datasets, models, and tests.
  • New: A model's summary view has been redesigned to summarize further important informational points about the model (for example, most important features). To learn more, see View a model's summary.
  • New: H2O Model Validation offers new demo models, datasets, and tests. To learn more, see Turn on or off demo mode

Tests

  • New: H2O Model Validation now supports the following tests:
  • New: Now, you can generate insights for a validation test after defining a large language model (LLM) source.
  • New: Now, H2O Model Validation manages all validation tests in the Tests card. In particular, you can now conduct the following validation test actions in the Tests card:
    • Create one of the supported validation tests
    • Create a validation test with the settings of a created test
    • Abort a validation test
    • View a test
    • Rename a validation test
    • Compare validation tests
    • Download a validation test's logs
    • Delete a validation test
  • New: Now, H2O Model Validation lets you access a test's logs when viewing it. To learn more, see View a validation test
  • New: Now, H2O Model Validation in the tests table places the newest tests at the top.
  • Improvement: Now, when defining the settings for a backtesting test, H2O Model Validation displays the model's forecast horizon period (the period that the test data spans). To learn more, see Model forecast horizon
  • Improvements: The chart (metric) titles for an adversarial similarity test have been renamed.
  • Improvements: The charts (metrics) for the following tests have been improved: Size dependency and backtesting.

Datasets

  • New: Now, H2O Model Validation does not allow you to delete datasets linked to an existing test.
  • New: Now, H2O Model Validation in the datasets table places the newest datasets at the top.
  • New: Now, you can generate insights for a dataset after defining a large language model (LLM) source.

Models

  • New: Now, when importing a model to H2O Model Validation, if you have already imported the model, H2O Model Validation alerts (flags) the model as imported.
  • New: Now, H2O Model Validation does not allow you to delete models linked to an existing test.
  • New: Now, H2O Model Validation in the models table places the newest models at the top.
  • New: Now, you can generate insights for a model after defining a large language model (LLM) source.

Connections

  • New: H2O Model Validation now supports Connections to My AI Engine Manager. To learn more, see Create a Connection

Documentation

  • New: All new features and settings for v0.16.0 have been documented
  • New: A new UI and UX for the documentation website are now available
Feedback