Improve Model Fine Tuning and Release Process on Cloud Native Environment

[rootfs]

Is your feature request related to a problem? Please describe.
The current (Modern)BERT classification models are trained by scripts and are expected to be manually executed. The fine tuned models are also manually uploaded to HuggingFace.

For Cloud Native environment, the fine tuning and model management are expected to be declarative, versioned, and provenanced, subject to the quality and safety control by the model providers.

Describe the solution you'd like

• Additional manifests, scripts, and automation framework to consume model training, downloading, and loading.
• Support model training and governance frameworks such as Kubeflow, AWS Bedrock, etc.