PIP Package for Model Registry

Overview

As a datascientist , we would like to provide the functionality of uploading via a python client / pip package

Functionalities

For the initial iteration of pip package aimed at facilitating a model registry feature, we would aim for the following functionalities :

Model Upload: Capability to upload machine learning models to the registry from local files or directories. This should include support for various model formats (e.g., TensorFlow SavedModel, PyTorch models, ONNX models) and associated artifacts.

Model Versioning: Ability to manage multiple versions of the same model within the registry. This involves assigning version numbers to uploaded models and maintaining a history of changes for each version.

Model Retrieval: Functionality to retrieve models from the registry based on specified criteria, such as model name, version, tags, or metadata attributes. This includes options for downloading models locally or accessing them programmatically for inference or further development.

Access Control: Mechanisms for controlling access to the models stored in the registry, ensuring that only authorized users or applications can upload, retrieve, or modify models. This may involve role-based access control (RBAC), permissions management, and authentication mechanisms.

Implementation details:

More details of Package Structure to be flushed out

cc @kbychu