Enzyme: Cross-language Automatic differentiation for LLVM IR

csail.mit.edu

30 pointsdhash3 comments

Open articleView on HN

Hi all, author here.

A couple of relevant links for the curious

Github: https://github.com/wsmoses/Enzyme

Paper: https://proceedings.neurips.cc/paper/2020/file/9332c513ef44b...

Project: enzyme.mit.edu

Basically the long story short is that Enzyme has a couple of really interesting contributions:

1) Low-level AD IS possible and can be high performance

2) By working at LLVM we get cross-language and cross-platform AD

3) Working at the LLVM level actually can give more speedups (since it's able to be performed after optimization)

4) We made a plugin for PyTorch/TF that uses Enzyme to import foreign code into those frameworks with ease!

Hello, is it a forward or a backward (or both) differentiation library?

How does the performnce of your code compare to Julia's autodiff source to source code? Does it solve from the exp/sin/cos problem? (By differentiating a polynomial approximation you use an approximation one degree lower)

It looks like the links to the code have some junk characters at the end. It should be https://github.com/wsmoses/Enzyme.

Reply on news.ycombinator.com