Customization on top of pytorch-lightning for quick and easy experimentations. Inspired by
- Speechbrain's mixed yaml configuration, dependency injection and code customization approach.
- Pytorch-lightning's Modular approach.
- AWS Cloudformation's Infrastucture-as-code
- Plug-and-playable loss functions, models, data-modules and loggers.
- Configuration-as-code
- Pass by name, minimal pass by position
- Functional and high composability
- Dependency injection
- Dataloader design is not as plug-and-play as expected
- Grouped batching based on data length.
- Think of a shorter name
- Enable multi-gpu processing