Skip to content


Repository files navigation


Deep Learning with PyTorch made easy 🚀 !

v0.5.x WIP!

Here are the main design principles:

  • The codes should be 'module first', which means all previous models should be a simple module now.
    • And model should only be related to the training stuffs. If we only want to use the fancy AI models at inference stage, module should be all we need.
  • The modules should be as 'native' as possible: no inheritance from base classes except nn.Module should be the best, and previous inheritance-based features should be achieved by dependency injection.
    • This helps the modules to be more torch.compile friendly.
  • Training stuffs are not considered at the first place, but they will definitely be added later on, based on the modern AI developments.
  • APIs will be as BC as possible.


carefree-learn is MIT licensed, as found in the LICENSE file.