Clear, intuitive syntax that lets you focus on the model and leave performance to the framework.
Mix-and-match inference methods, proposers, and inference strategies to achieve maximum efficiency.
Powered by PyTorch
Leverage native GPU and autograd support and integrate seamlessly with the PyTorch ecosystem.
Status: Beta. APIs are likely to change. Functionalities are constantly being improved. Bug reports are welcome, but bandwidth is very limited for feature requests.