T3W is a lightweight framework for training PyTorch models written by Yuyao Huang during his PhD at Tongji University.

• T3W is "typed". It leverages stronger and static typing compared to normal python code for clearer architecture and less bugs. The programming model is object-oriented. Users (you) are required to implement interfaces as subclasses and inject them as dependencies.
• T3W is "thin". With the philosophy "less is more" in mind, it leverages a minimal self-contained codebase that basically only depends on PyTorch to run. The plugin system under interface ISideEffect makes T3W not only thin, but also highly extensible.
• T3W stands with "PyTorch".

See the concise example mnist_example.py.

If you feel like using t3w.py, you can install it with pip install t3w.

API documentation is currently available at https://tjyuyao.github.io/t3w/api/. We are going to add more detailed user guide in the future.