Add the MLOps platform as part of the GenAI infra

[andreeamun]

Description

Add MLOps platform to the proposed infrastructure to enable organisations to automate their ML workloads.

Issues

List the issue or RFC link this PR is working on. If there is no such link, please mark it as n/a.

Type of change

Dependencies

• Have a K8s cluster underneath.

Tests

Describe the tests that you ran to verify your changes.

[mkbhanda]

@andreeamun thank you for your PR. Would like to understand better how MLOPS fits in currently with OPEA, which is focused on RAG/GenAI pipelines. Also the Data science kit. I do agree they both pertain to machine learning but need some use cases to justify why we want to pull it in here. This might be a logical evolution of OPEA and we may also pull in Jupyter Notebooks. Further, if are bringing in Ubuntu related materials, there will need to be more documentation. Currently we have not made any requirements on the operating system running on the platforms/nodes.

[andreeamun]

thank you @mkbhanda . An MLOps platform such as Charmed Kubeflow is crucial when it comes to RAG / GenAI pipelines because it automates some of the workloads and enables easy iteration. To be more precise, MLOps platforms can be used to:

• Automate pipelines used for data ingestion
• Perform model optimisation (eg: hypterparameter tuning, fine-tuning, p-tuning)
• Train models and automate some of those pipelines to efficentise work
• Benefit from user management, network isolation and further security enhancements for larger teams
• Model registry / Experiment tracking with MLflow

Such a platform is a cloud-native application which runs on any CNCF-conformant Kubernetes, giving enterprises freedom to build on their existing infrastructure, regardless of the OS running underneath. However, I am happy to provide more details on what are the benefits of using Ubuntu as the recommended operating system.

Please let me know if this answers your question.

[daisy-ycguo]

@andreeamun Thank you for the contribution. Will you rebase your PR and sign off your PR? Refer to the guidance here: https://github.com/opea-project/GenAIInfra/pull/326/checks?check_run_id=29702935232

[KfreeZ]

thank you @mkbhanda . An MLOps platform such as Charmed Kubeflow is crucial when it comes to RAG / GenAI pipelines because it automates some of the workloads and enables easy iteration. To be more precise, MLOps platforms can be used to:

• Automate pipelines used for data ingestion
• Perform model optimisation (eg: hypterparameter tuning, fine-tuning, p-tuning)
• Train models and automate some of those pipelines to efficentise work
• Benefit from user management, network isolation and further security enhancements for larger teams
• Model registry / Experiment tracking with MLflow

Such a platform is a cloud-native application which runs on any CNCF-conformant Kubernetes, giving enterprises freedom to build on their existing infrastructure, regardless of the OS running underneath. However, I am happy to provide more details on what are the benefits of using Ubuntu as the recommended operating system.

Please let me know if this answers your question.

If we want to support the kubeflow or DSS, we'd better provide some detailed use case, examples and code changes, scripts whatever needed.
When we can see the OPEA examples are running on top of kubeflow or DSS, then we can document that we support it.

[mkbhanda]

thank you @mkbhanda . An MLOps platform such as Charmed Kubeflow is crucial when it comes to RAG / GenAI pipelines because it automates some of the workloads and enables easy iteration. To be more precise, MLOps platforms can be used to:

• Automate pipelines used for data ingestion
• Perform model optimisation (eg: hypterparameter tuning, fine-tuning, p-tuning)
• Train models and automate some of those pipelines to efficentise work
• Benefit from user management, network isolation and further security enhancements for larger teams
• Model registry / Experiment tracking with MLflow

Such a platform is a cloud-native application which runs on any CNCF-conformant Kubernetes, giving enterprises freedom to build on their existing infrastructure, regardless of the OS running underneath. However, I am happy to provide more details on what are the benefits of using Ubuntu as the recommended operating system.
Please let me know if this answers your question.

If we want to support the kubeflow or DSS, we'd better provide some detailed use case, examples and code changes, scripts whatever needed. When we can see the OPEA examples are running on top of kubeflow or DSS, then we can document that we support it.

Thank you @KfreeZ for your input! @andreeamun OPEA is starting from solving end-user problems with reference implementations. Perhaps a use case illustrated using in GenAIExamples of fine-tuning or data-prep would be more compelling then just documenting how to install Kubeflow. We have RFCs as PRs in https://github.com/opea-project/docs where we can collaborate on any use case you choose to highlight value.