This repository provides a framework to experiment with different deep learning models with Pytorch, streamlining training, evaluation, and generation (for generative models). BEM is designed to satisfy the following properties:
- Flexibility: Allows easy integration of custom models and methods.
- Modularity: Components like data handling, training, evaluation, and logging are modular.
- Extensibility: By implementing the required functions, you can support your own generative models.
- Reproducibility: Saving experiment parameters and metrics to ensure reproducibility and organized storage.
- Device Support: Utilizes available hardware (CPU, CUDA, MPS).
This repo provides a unique framework to experiment with different deep generative models, making it vastly easier to:
- Train them
- Evaluate them
- Compare them
See our how_to_use_bem.ipynb
notebook.