[PROPOSAL] ml-commons Console OSD plugin

[wanglam]

What/Why

What are you proposing?

We are proposing an administration UI for the machine learning library ml-commons to make some of its features visualized. We are going to build a console panel to manage the trained models and uploaded models as well as committed tasks.

What problems are you trying to solve?

Managing the trained model using only API is painful. According to current API, a model ID will be responded to the user and he need to record it down manually. And once the model ID is forgotten, it will be super hard for the user to recall this model. Thus it is quite necessary to build an UI to enable the users to browse / filter all the trained models. The model filtering will also enable the user to dispose those useless models, once the models are listed, they can be deleted just by one click.

Targeting for 2.4, the ml-commons will have a release to support custom deep learning model deployment. Therefore a new problem has been raised that where should the user upload their own model file. This console also plan to optionally support model uploading from the UI, bringing some connivence to the user experience.

What is the user experience going to be?

For model management, we are going to built several pages for model management:

1. A page listing all the models trained in the history, with the feature that the users can filter the models by algorithm, training time and the model hyper-parameters. The user can delete the useless model on this page
2. A model detail page containing all the properties of the model as well as the corresponding tasks and logs.
3. A task listing page let the user to mange the tasks. The user can terminate some going tasks that they don’t want to run at this moment.
4. A custom model listing page for the management of uploaded model files, where users can enable / disable / delete the models.
5. A model upload page where the user can drag-n-drop their model file and upload to OpenSearch.

Why should it be built? Any reason not to?

N/A

What will it take to execute?

The project is mostly a front end project, thus FEEs are required. Based on the current scope (training UI for 5 algorithms + model management UI, might be changed after the final design), the estimated workload is about 30 man*day. we need some effort on UX so as to 1) define the accurate user workflow as well as to organize the concrete user interactions. 2) design the visual layout of model training and management UI. Some minor requirements to the backends are also observed, the SDEs should do some upgrade to the ml-commons API.