Workaround to Deploy Trained Package when Training Completes but Pipeline Fails

News Knowledge Base
,
1

Is there a way to still deploy a model if training shows as completed in the logs, but the training pipeline fails?

Issue:

The pipeline shows as finished/completed in the pipeline logs and artifacts are visible in the Output of the pipeline, however, the pipeline status shows as Failed and the creation of a new version for the model did not complete.

If this issue occurs, a ticket should be raised with the UiPath Product Support team for investigation as to why the new version of the model failed to be created.

While investigation is ongoing to determine why the pipeline failed, the following steps may be taken to recover the trained model if the required files are found in the output of the pipeline and the pipeline shows as completed in the pipeline logs.

Workaround Steps:

  1. Go to the Pipeline and download the zip file ".zip“ . Save this for step 4.
image.png

  1. Go to ML Packages -> Package Used For the Training -> Search for the latest version of the package under the versions tab. Download the artifacts. Only the “…metadata.json“ file will be needed from the download. If no previous versions of the model have been trained and only minor version 0 is available, reach out to the product team to request the metadata.json file for the base package of the model to complete the steps for this workaround.

  1. Open the metadata json file and find the “trainingVersion” field. Increase the number by 1. For example, change from 1 to 2. Save the file under a different name
  2. In the ML Packages view go to Import ML Package. Use the zip file from the pipeline and the edited metadata file to upload a new package

  1. A new version should be shown under the original package