New Owl book: Architecture of Advanced Numerical Analysis Systems

There is a second book on Owl: Architecture of Advanced Numerical Analysis Systems - Designing a Scientific Computing System using OCaml.

Free download in PDF and ePub formats!


A post was split to a new topic: Is it possible to use JuliaLang as a ‘backend’ for OWL?

I have been reading this book for some time now. It’s an incredible book, a gem! I haven’t seen anything like it on the topic before.

1 Like

Self-advertisement time: I believe that the Open Access nature of the book (the free access to the online version) was made possible by a grant from the OCaml Software Foundation.

Previously we funded John Whitington to make his own book OCaml from the Very Beginning available online for free as well. In general the Foundation is interested in improving existing online content about OCaml but also making new valuable content available.


Thanks to everyone involved in making this work freely available!


What is the thing you use to target GPUs?
Just save your model in a specific format, then use some python code
to load that model and run it on a GPU?

1 Like

You mean in owl? That would be onnx generated via owl_symbolic. Not sure if it is still maintained (ping @jrzhao42)

What is the largest Neural Network that has been trained from scratch using OWL?

I’m not referring to downloading a pretrained model and doing inference, but training from scratch. I also realize most of the current popular techniques involves downloading a large pretrained model then tweaking it.

I’m confident people have trained MNIST / CIFAR recognizers, but has anyone trained an ImageNet recognizer (or something larger?) from scratch using OWL ?

I found this book while I was looking if owl supported gpu. Thanks for this book!

I have to read “OCaml Scientific Computing: Functional Programming in Data Science and Artificial Intelligence” first ? Interested to read about SIMD and Vectorization.

Thanks for the reference! Unfortunately that repo is not maintained anymore, but the core idea is still relevant.