Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Keras Core: Keras for TensorFlow, Jax, and PyTorch

keras.io

Post reply on HN
Print view View original

1–10 of 73 posts

Re: Keras Core: Keras for TensorFlow, Jax, and PyTorch

#2

Post by dewitt »

From the announcement:

"We're excited to share with you a new library called Keras Core, a preview version of the future of Keras. In Fall 2023, this library will become Keras 3.0. Keras Core is a full rewrite of the Keras codebase that rebases it on top of a modular backend architecture. It makes it possible to run Keras workflows on top of arbitrary frameworks — starting with TensorFlow, JAX, and PyTorch."

Excited about this one. Please let us know if you have any questions.

Re: Keras Core: Keras for TensorFlow, Jax, and PyTorch

#4

Post by Narew »

Keras was already that some years ago. It supported tensorflow, theano, mxnet if my memory is right. And then they ditched everything for tensorflow. At the time it was really hard to use keras without calling backend directly for lots of optimisation, unsupported feature on they API etc... This make the use of Keras not agnostic at all. What's different now ?

Re: Keras Core: Keras for TensorFlow, Jax, and PyTorch

#5

Post by minimaxir »

Narew wrote: post #4

Keras was already that some years ago. It supported tensorflow, theano, mxnet if my memory is right. And then they ditched everything for tensorflow. At the time it was really hard to use keras without calling backend directly for lots of optimisation, unsupported feature on they API etc... This make the use of Keras not agnostic at all. What's different now ?

> What's different now ?

PyTorch adoption: back when Keras went hard into TensorFlow in 2018, both TF and PyTorch adoption were about the same with TF having a bit more popularity. Now, most of the papers and models released are PyTorch-first.

Re: Keras Core: Keras for TensorFlow, Jax, and PyTorch

#7

Post by kerasteam »

syntaxing wrote: post #6

Does this mean the weights output can be backend agnostic? Also, are there any examples using this for the coral TPU?

Yes, model weights saved with Keras Core are backend-agnostic. You can train a model in one backend and reload it in another.

Coral TPU could be used with Keras Core, but via the TensorFlow backend only.

Re: Keras Core: Keras for TensorFlow, Jax, and PyTorch

#9

Post by sbrother »

kerasteam wrote: post #3

I worked on the project, happy to answer any questions!

This looks awesome; I was a big fan of Keras back when it had pluggable backends and a much cleaner API than Tensorflow.

Fast forward to now, and my biggest pain point is that all the new models are released on PyTorch, but the PyTorch serving story is still far behind TF Serving. Can this help convert a PyTorch model into a servable SavedModel?

Post reply on HN