Migrate Codebase to Latest PyTorch Lightning and NumPy

[ruddyscent]

Description:

This PR updates the project to be compatible with the latest version of PyTorch Lightning and NumPy. The key changes include:

1. Remove Deprecated compute_on_step Argument from Metrics:

   • Updated metric classes to remove the deprecated compute_on_step argument, ensuring they work correctly with the latest TorchMetrics and PyTorch Lightning.
   • Refactored metric initialization and handling to follow current best practices.

2. Replace Deprecated np.int Usage:

   • Replaced the deprecated use of np.int with the appropriate integer type, ensuring compatibility with newer versions of NumPy.

3. Update Hyperparameter Handling in TrainingModule:

   • Refactored the TrainingModule class to use self.save_hyperparameters() for managing hyperparameters, replacing the direct assignment of self.hparams. This aligns with the latest PyTorch Lightning guidelines and prevents potential errors.

4. Migrate Classes to the Latest PyTorch Lightning:

   • Updated several classes and methods to be compatible with the latest PyTorch Lightning API, ensuring smooth operation and future-proofing the codebase.

5. Add requirements.txt to Specify Project Dependencies:

   • Introduced a requirements.txt file to clearly define and manage the project's dependencies.

Testing:

These changes have been tested and confirmed to work within containers based on the nvcr.io/nvidia/pytorch:24.07-py3 image. This ensures compatibility and smooth integration in the specified environment.