An abridged collection of CUDA and Pytorch C/C++ extensions. Each provides some basic transformation on GPU data as part of an ML model's data flow. Many make use of the Torch C++ API for easy CMake compilation and incorporation into a PyTorch model training loop. Some older libraries use the Ninja compiler built into python's buildtools.
Given that these libraries were each a small part of some architecture concept under rapid investigation and prototyping, they include no tests.