Tensor Comprehensions in PyTorch

Tensor Comprehensions (TC) is a tool that lowers the barrier for writing high-performance code. It generates GPU code from a simple high-level language and autotunes the code for specific input sizes. We highly recommend reading the Tensor Comprehensions blogpost first. If you ran into any of the following scenarios, TC is a useful tool for you. Tensor Comprehensions are seamless to use in PyTorch, interoperating with PyTorch Tensors and nn Variables.