UiPath Orchestrator

ML Skills

An ML skill is a deployed, consumer-ready ML or OS package. Once deployed as ML skills, both ML Packages and OS Packages become models ready to be consumed within RPA workflows.

When you are deploying an ML or OS package into an ML Skill, the model .zip file corresponding to the package is extracted and containerized on AI Fabric's Kubernetes cluster, on the AI Fabric server. The container exposes a REST endpoint so that it can be called into an RPA Workflow using the ML Skill activity. To be able to consume a model in your Studio workflows you have to install the UiPath.MLServices.Activities package. This activity package is only available for Studio v2019.10+ and can only be used by Robots v2019.10+. Read more about it here.

It is recommended that users acting as Process Controllers handle model deployments. More details about recommended user personas here.

The ML Skills page displays all the models deployed on your Orchestrator tenant, whether they use ML or OS packages. ML skills represent a global resource, meaning that they are available across all folders, irrespective of their types.

The ML Skills page enables you to view all the ML or OS packages deployed, along with their statuses, versions, whether or not a GPU is required, predictions, and descriptions. Moreover, you can deploy packages by creating ML skills, delete them, view their detailed information, parameters, versions, or manage package versions - if available.

Creating ML Skills

Note:

You cannot deploy more than 2 ML skills per AI Robot License.

  1. In the ML Skills page, click Add. The ML Skills > Create page is displayed, enabling you to deploy a model based on ML or OS packages.
  2. In the ML Skills > Create page, select the desired type of package:
    • ML Packages - enables you to create an ML skill that uses an ML package uploaded on your tenant by one of your data scientists.
    • OS Packages - enables you to create an ML skill that uses an OS package created by UiPath ML engineers.
      Continue with the steps corresponding to your selected package type.
  1. On the Create ML Skill window, click ML Packages or OS Packages. One of the following windows is displayed, according to your choice:
    • Create ML Skill from ML Package - enables you to select an ML package uploaded on your Orchestrator tenant to create a new ML skill
    • Create ML Skill from ML Package - enables you to select an OS package available in AI Fabric to create a new ML skill.
  2. Select a package from the ML Package Name drop-down. You can select ML packages in the Create ML Skill from ML Package window, respectively OS packages in the Create ML Skill from OS Package window. The selected package's details (description, version, changelog for ML packages, input and output descriptions, and when it was published) are displayed in a panel on the right of the page.
  3. Select a package version from the ML Package Version drop-down.
  4. Fill in a name on the ML Skill Name field to easily identify the skill later on.
  5. (Optionally) Add a description to the Description field.
  6. Choose whether or not to enable a GPU on the environment running this skill.
  7. Click Create. The Create ML Skill page is closed and the ML Skills page is displayed. The model is wrapped in UiPath's serving framework and deployed within a namespace on AI Fabric's Kubernetes cluster that is only accessible by your Orchestrator tenant.

Note:

Model deployment takes up to 10-15 minutes.

If the deployment is successful, the status of the ML skill changes from Deploying to Available.

Managing Package Versions

In the ML Skills page, click the More Actions button next to a deployed skill and select Details. The Package Versions window is displayed. The Version Management tab within the window displays information about the ML package from which the skill was derived, its status, and a list of versions of the package.
To enable or disable the usage of GPUs for the currently used package version use the Change button. Perform the desired GPU configuration change in the Update Current ML Skill window, then click Deploy to save the changes.

There are three ways of handling package version management:

Using a Specific Package Version

  1. On the Version Management tab, click Use for the version you want to use. The button is disabled for the version that is currently in use, marked as Current. The Upgrade Current ML Skill window is displayed with details about the ML Skill and the possibility to enable/disable GPU.
  1. In the Upgrade Current ML Skill window, enable or disable GPU, and click Upgrade to save the selected configuration.

Note:

You can enable or disable the usage of GPUs for the currently used package version using the Change button. Perform the desired GPU configuration change in the Update Current ML Skill window, then click Deploy to save the changes.

Updating to the Latest Package Version

  1. On the Version Management tab, click the Latest button if you want to update the skill to the latest available package version. The Package Versions window is closed and the Upgrade Current ML Skill window is displayed with details about the ML Skill and the possibility to enable/disable GPU.
  1. In the Upgrade Current ML Skill window, enable or disable GPU, and click Upgrade to save the selected configuration.

Rolling Back to the Previously Used Package Version

  1. On the Version Management tab, click the Rollback button if you want to return to the previously used version. The Package Versions window is closed and the Rollback ML Skill window is displayed with details about the ML Skill and the possibility to enable/disable GPU.
  1. In the Rollback ML Skill window, enable or disable GPU, and click Rollback to perform the configuration change.

Removing ML Skills

  1. In the ML Skills page, click the More Actions button next to a deployed skill and select Remove. A confirmation window is displayed.
  2. Click Yes to remove the skill. The selected skill is undeployed and it disappears from the ML Skills page.

Updated 2 days ago



ML Skills


Suggested Edits are limited on API Reference Pages

You can only suggest edits to Markdown body content, but not to the API spec.