Skip to main content
Version: v1.2.X

Release notes

v1.2.22 | September 17, 2026​

H2O AI Managed Cloud v26.10.0 includes H2O Label Genie v1.2.22.

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.21 | September 13, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.
  • Fixed zero-shot predictions in text-generative AI tasks stopping the whole batch when a row and Max response tokens do not fit in the model's context window, or when Max response tokens exceeds the model's limit. Such predictions now cut the end of the row's longest column values, lower the response token limit, or both, without being marked.
  • Fixed uploads of a dataset, a label file, or annotation instructions failing when the filename contains '#' or '?'.
  • Fixed H2O Label Genie v1.2.20 failing to start on hosts with FIPS mode turned on.

v1.2.20 | August 14, 2026​

New features​

  • In image classification and image regression tasks, you can now click the image to open a larger view of it.
  • The container image is now smaller.

Fixes​

  • Resolved security vulnerabilities to improve app stability.
  • Fixed background jobs (for example, zero-shot predictions) lowering the CPU priority of H2O Label Genie itself, which could make H2O Label Genie stop responding for minutes.
  • Fixed an issue with the Annotate tab of a classification task that has one class, which in v1.2.19 crashed when an unapproved sample had a zero-shot prediction.
  • Fixed an issue with imports of zip files created by macOS Finder or a similar tool not keeping non-ASCII characters (for example, 'ö') in the filenames and folder names inside them.
  • Fixed imports of an empty CSV or Parquet label file, or of a Parquet label file that cannot be read, crashing instead of showing an error message.
  • Fixed the google/pegasus-large zero-shot model for text summarization failing to load in v1.2.16 to v1.2.19, and in air-gapped deployments.
  • Fixed a sample showing the image or audio file of another sample in the same dataset when images in different folders have the same filename (for example, 'train/001.jpg' and 'val/001.jpg'), or when audio filenames match up to their first period (for example, '001.wav' and '001.flac').

v1.2.19 | July 27, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.18 | June 26, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.
  • Fixed imports of a text dataset that is not UTF-8 encoded crashing instead of showing an error message that asks for a UTF-8 file.
  • Fixed text dataset imports from AWS S3 or H2O Drive crashing instead of showing an error message when the file cannot be downloaded.

v1.2.17 | May 25, 2026​

New features​

  • The container image can now run under an arbitrary user ID.

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.16 | May 19, 2026​

New features​

  • The container image now uses CUDA 13 and PyTorch 2.11 and is smaller.
  • GPU deployments now need an NVIDIA R580 or higher driver; with an earlier driver, models run on the CPU.

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.15 | May 8, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.14 | April 25, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.13 | April 17, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.12 | April 7, 2026​

Fixes​

  • Resolved security vulnerabilities to improve app stability.

v1.2.11 | March 20, 2026​

New features​

  • The Annotate and Review tabs now show an identifier for each sample, which you can hide by turning off Show sample identifier in Settings.
  • H2O Label Genie can now download models from H2O ModelHub deployments that require authentication.
  • The app now logs its main actions (for example, imports, exports, and background jobs) to help support teams resolve issues.
  • When it runs in H2O AI Cloud, H2O Label Genie now finds H2O ModelHub automatically and downloads models from it.
  • Clustering results now load faster, especially for large datasets.
  • The container image now uses a FIPS-enabled base image.

Fixes​

  • Resolved security vulnerabilities to improve app stability.
  • Fixed Gaussian-Mixture clustering, the default Clustering model, sometimes producing clusters close to random.
  • Fixed clustering tasks failing with a 404 error in air-gapped deployments when they download their model from H2O ModelHub.
  • Fixed Cancel not ending the selection after you click Delete clustering task(s) on the Data exploration page.

v1.2.0 | March 19, 2025​

Fixes​

  • Resolved security vulnerabilities to improve app stability.
Feedback