[Spike] Investigate creation of a registry stack for AI development

[thepetk]

Which area this issue is related to?

/area registry

Issue description

As an application developer or an ML engineer who wants to develop AI solutions I'd like to have a starting point where I can develop ML/AI applications within the kubernetes and OpenShift Environments.

To achieve that, a new registry stack can be created including all the necessary ML & AI tools that an ML engineer should have to work on their ML or AI applications.

This issue focuses on the investigation process around the technologies & tools that this registry stack should include.

Acceptance Criteria

• [ ] The contents of the registry stack are defined.

[michael-valdron]

Investigate Eclipse Che requested stack for llama

[michael-valdron]

We can start out with the following frameworks:

• PyTorch
• Tensorflow v2
• FastAI

[thepetk]

We can start out with the following frameworks:

• PyTorch
• Tensorflow v2
• FastAI

I was wondering if it makes sense to have 3 separate stacks? (e.g. following the python-flask / python-django separation).

[michael-valdron]

We can start out with the following frameworks:

• PyTorch
• Tensorflow v2
• FastAI

I was wondering if it makes sense to have 3 separate stacks? (e.g. following the python-flask / python-django separation).

We decided in a previous team discussion to use one stack with three starter projects (one for each framework). We also deciding to use a purpose built image with this stack to provide common dependencies rather than placing everything in each starter projects dependency definitions.

[michael-valdron]

Reviewing current revision: https://docs.google.com/document/d/1oolCZRL4eAiuqTufMni6sOiTQjH7lsfwD09l_BU-G2A/edit?usp=sharing