4 ms·

I'm really excited about this project. Seeing Elixir inch this much closer to being a viable option for machine learning and related fields is great. In case s

by juhatl 6y ago

I'm really excited about this project. Seeing Elixir inch this much closer to being a viable option for machine learning and related fields is great. In case someone here wants to read rather than listen to all the details, I wrote down some notes over at: https://elixirforum.com/t/anyone-who-wants-to-speculate-about-this-tweet-from-jose/35772/69?u=juhalehtonen https://elixirforum.com/t/anyone-who-wants-to-speculate-abou...

flyingfences 6y ago
Thanks, that actually explains it pretty well: > Nx = Project for a collection of libraries. Nx is the core library, the other libraries depend on this core library > If you come from Python, it can be though of as kind of like Numpy. Long way to go but working on that. > “Slowness in Elixir” due to immutability and copying etc. Performance comes from module that is part of Nx called Numerical Definitions > In the definition you write with a subset of Elixir. Elixir kernel is replaced with Numerical kernel. > Based on Google XLA (accelerated linear algebra) > You can give it a computation graph and exla compiles it to run efficiently on CPU or GPU
- BadInformatics 6y ago
  Pretty disappointing to see it using XLA. Anything Tensorflow is notoriously difficult to build on a local machine and generally considered a dependency nightmare. I would've thought libtorch bindings would be much easier.
  - josevalim 6y ago
    Our compiler backends are pluggable. We went with XLA because that seemed the most accessible to us 3 months ago but you should be able to bring in libtorch or any other tensor compiler. That's actually one of the things I am really looking forward to: see what other compilers people will integrate with. I am aware some of the neural network libs have pluggable compilers but it will be interesting to see it done at the tensor level. :)
    - BadInformatics 6y ago
      That's great to hear. I'd love to see the Bazel monstrosity that is Tensorflow XLA get easier to build because of Nx. Not holding my breath though!
      - josevalim 6y ago
        Yeah, I would love if we could depend only on XLA. I think other communities like Julia could benefit from it too.
- dynamite-ready 6y ago
  So, two questions. Is this a standard library module, and does this run on the GPU?

by juhatl 6y ago

flyingfences 6y ago

Thanks, that actually explains it pretty well: > Nx = Project for a collection of libraries. Nx is the core library, the other libraries depend on this core library > If you come from Python, it can be though of as kind of like Numpy. Long way to go but working on that. > “Slowness in Elixir” due to immutability and copying etc. Performance comes from module that is part of Nx called Numerical Definitions > In the definition you write with a subset of Elixir. Elixir kernel is replaced with Numerical kernel. > Based on Google XLA (accelerated linear algebra) > You can give it a computation graph and exla compiles it to run efficiently on CPU or GPU

BadInformatics 6y ago
Pretty disappointing to see it using XLA. Anything Tensorflow is notoriously difficult to build on a local machine and generally considered a dependency nightmare. I would've thought libtorch bindings would be much easier.
- josevalim 6y ago
  Our compiler backends are pluggable. We went with XLA because that seemed the most accessible to us 3 months ago but you should be able to bring in libtorch or any other tensor compiler. That's actually one of the things I am really looking forward to: see what other compilers people will integrate with. I am aware some of the neural network libs have pluggable compilers but it will be interesting to see it done at the tensor level. :)
  - BadInformatics 6y ago
    That's great to hear. I'd love to see the Bazel monstrosity that is Tensorflow XLA get easier to build because of Nx. Not holding my breath though!
    - josevalim 6y ago
      Yeah, I would love if we could depend only on XLA. I think other communities like Julia could benefit from it too.
dynamite-ready 6y ago
So, two questions. Is this a standard library module, and does this run on the GPU?