> why not Pytorch? JAX enables using (parts of) existing numpy codebases in disciplines other than deep learning. Autodiff and compilation to GPUs are very use

> why not Pytorch? JAX enables using (parts of) existing numpy codebases in disciplines other than deep learning. Autodiff and compilation to GPUs are very useful for all kinds of algorithms and processing pipelines.