Re:Deep Learning with PyTorch: A 60 Minute Blitz [video]
by ibab on Sunday October 13, 2019 @09:50AM
(#21240333)
I really like JAX as well: https://github.com/google/jax.
It's younger than PyTorch and TF, but feels cleaner and more expressive. It has a very nice autodiff implementation (based on https://github.com/HIPS/autograd) and performance is comparable to TF in my experience.
[ Parent
| on HN ]
Re:Deep Learning with PyTorch: A 60 Minute Blitz [video]
by solidasparagus on Sunday October 13, 2019 @02:37PM
(#21241997)
It feels like JAX doesn't have any of the high-level APIs that PT/TF/MXNet that are vital for fast prototyping of model architectures. Is that correct?
Re:Deep Learning with PyTorch: A 60 Minute Blitz [video]
by ibab on Sunday October 13, 2019 @03:16PM
(#21242282)
It has stax, which is a minimal example of how to build a high level library: https://github.com/google/jax/blob/master/jax/experimental/s...
It seems that the JAX developers are focusing their time on making the core framework better and are leaving the task of building high-level APIs to the community for now. I suspect we'll see a few high-level APIs emerge over the next few months that explore different approaches before the community settles on a particular one.
Re:Deep Learning with PyTorch: A 60 Minute Blitz [video]
by solidasparagus on Sunday October 13, 2019 @03:25PM
(#21242348)
I hope not. That's part of what makes TF so miserable - the core library didn't provide the tooling people actually needed so the community built a ton of different tools and it just made TF confusing to use.