A machine learning team is developing a project that involves integrating a custom PyTorch model into their Databricks ML pipeline. They aim to make the PyTorch library and the custom model accessible for training across all notebooks within the workspace. What is the best practice to achieve this?