Create a worker on Arkindex

You can publish an existing worker to Arkindex following the documentation for worker developers, but you can also create workers and worker versions from the Arkindex frontend.

Warning

To be able to create workers, and worker versions, you need a dedicated user permission. If you are unable to create workers and worker versions, contact an instance administrator.

Create a worker

To create a new worker on Arkindex, first click on Workers in the user e-mail dropdown menu in the top-right corner.

Browse the workers you have access to

Then click the Create button above the list of workers on the right.

List of workers and create button

This opens the worker creation modal. You need to fill out the following fields:

• Worker name. Usually, this contains the name of the technology used (like Doc UFCN, PyLaia, SpaCy etc) or a description of what the worker does (such as Page Splitter, Form Classifier etc). If the worker is specifically tailored to a dataset, you may also include that dataset’s name in the worker name.
• Worker slug. Worker slugs are unique on an Arkindex instance.
• Worker type. This field is not limited to a pre-set list of worker types. However, you should use a meaningful type in relation to what the worker does (such as recognizer, dla etc) as you can use this type to filter your workers in the workers list.

You can then click the Create button at the bottom of the modal.

New worker creation modal

A worker is only the parent object: when you run a process on Arkindex, you are using worker versions, which are specific versions of a worker. Worker versions of the same worker are based on the same technology or algorithm, but workers can evolve and improve and so whenever a worker is updated a new worker version is created. You can create worker versions for existing workers from the Arkindex frontend as well.

Create a worker version

Warning

To create a worker version on an existing worker, you need to have administrator rights on this worker.

Info

Worker versions can only be created from the Arkindex frontend for workers that are not linked to a Git repository. Workers created from the frontend as described above are not linked to a Git repository.

To create a new worker version for an existing worker, you need to select that worker in the workers list, then click on the Create button above the versions list.

Worker versions

This opens the worker version creation modal.

Worker version creation modal

• Docker image reference. If your worker is hosted on Teklia’s GitLab instance, this Docker image reference should look like this: registry.gitlab.teklia.com/my_group/my_worker:0.4.2. This is the reference to the Docker image containing your worker’s code. When you use a worker version in a process, Arkindex uses this Docker image to execute the worker. Without this Docker image reference, your worker version cannot be used in a process. This field is not required, because it is possible to create a worker version without a Docker image in the context of user worker runs.

• Configuration: this is where you should put all the required configuration parameters which would be in the YAML configuration if you were publishing your worker by following the documentation for worker developers. This is for example where you can define user configuration parameters, which users will then be able to set when using this worker version in a process. This field cannot be left empty. If you do not want to define any configuration parameters, you should leave the default value {}.

• Model usage: whether your worker requires, supports or does not support the use of a machine learning model. If your worker is a simple algorithm, such as one that splits images in half, then it likely does not require nor support the use of a model. More complex workers however can take data that is on Arkindex as input, feed it to a machine learning model, and then transform use the output results to create objects (elements, classifications, entities etc) on Arkindex. If your worker needs a model to work, then you should select required. If it can run with or without a model (perhaps there is a default model) then you can select supported. Otherwise, select disabled.

• GPU usage: whether your worker requires (required), supports (supported) or does not support (disabled) being run on GPUs.

Once the form is filled, click the Create button.

Your new worker version now appears in the worker version list, and can be used in processes.