Skip to content

OpenVINO Toolkit operator 1.1.0
Compare
Choose a tag to compare
@mzegla mzegla released this 08 Feb 10:40
· 17 commits to main since this release
v1.1.0
4e3ea7a
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Version 1.1.0 is a minor upgrade of operator 1.0.0.

It includes minor changes to example and default values of some CRD fields as well as some bug fixes.

It also enables users to configure metrics for single model mode from the CRD level.

ModelServer changes:

  • Added support for metrics configuration for single model mode. Previously metrics collection could be turned on only in the model server configuration file. New monitoring fields have been added to ModelServer CRD to enable setting metrics for deployments without the configuration file.

  • Updated default OVMS image to 2022.3 - the current latest available.

  • Fixed handling value 0 for file_system_poll_wait_seconds and sequence_cleaner_poll_wait_minutes parameters. Previously they were ineffective. Now providing 0 correctly disabling periodic model repository updates and stateful models sequence cleanup routines.

  • Changed the model server sample and the default values of model_path, plugin_config and file_system_poll_wait_seconds.

    • model_path is now generic GS path: gs://<bucket_name>/<model_dir>. Users need to specify their own path for the model server to successfully load the model.

    • plugin_config has been changed to {"PERFORMANCE_HINT":"LATENCY"} to match latest OpenVINO guides for latency-oriented setup.

    • file_system_poll_wait_seconds has been changed from 0 to 1.

  • Fixed typo in service type drop-down list option. Previously, in OCP form we had "Loadbalancer" available as a service type, which is wrong. Changed to "LoadBalancer".

  • Fixed handling update_strategy deployment parameter.

  • Fixed model server sample manifests. Changed model_config to config_path fields for multi model mode. model_config was wrong, and it caused deployment failure.

  • Changed liveness and readiness. Server is now alive when gRPC service is running and ready when all desired models are loaded. It is in line with the OpenVINO Model Server 2022.3 API changes.

  • Added metrics-based scaling with HPA example along with Prometheus setup.

Notebook changes:

  • Fix BuildConfig labels for RHODS integration. Added opendatahub.io/build_type: notebook_image label, so that RHODS dashboard gets notified about the build status.

  • Fix ImageStream annotations for RHODS integration. Removed opendatahub.io/notebook-python-dependencies annotation as we cannot guarantee dependencies versions and modified opendatahub.io/notebook-software annotation to drop dot version for Python (v3.8.6 -> v3.8).

Assets 2
Loading