malt

The framework used in the book, malt[0], is currently not GPU-accelerated, but it's being worked on. Maybe interesting, I used it for a toy implementation of the GPT architecture[1] in about 500 lines. (I studied with one of the authors, Dr. Daniel Friedman; wasn't super involved here but proofread a late draft and TA'd for a course based off the book.) [0]: https://github.com/themetaschemer/malt [1]: https://github.com/sporkl/malt-transformer

Malt: A Deep Learning Framework for Racket by Anurag Mendhekar and Daniel P. Friedman(“Lispman”) https://www.thelittlelearner.com/ > We discuss the design of a deep learning toolkit, Malt (https://github.com/themetaschemer/malt), that has been built for Racket. Originally designed to support the pedagogy of [The Little Learner—A Straight Line to Deep Learning](https://mitpress.mit.edu/9780262546379/the-little-learner/), it is used to build deep neural networks with a minimum of fuss using tools like higher-order automatic differentiation and rank polymorphism. The natural, functional style of AI programming that Malt enables can be extended to much larger, practical applications. We present a roadmap for how we hope to achieve this so that it can become a stepping stone to allow Lisp/Scheme/Racket to reclaim the crown of being the lan

Checking the library the book implements can help: https://github.com/themetaschemer/malt

I am looking at the GitHub repo for the Malt library developed in the book https://github.com/themetaschemer/malt. It looks like they use Racket (Scheme) vectors to implement tensors. I experimented with loading Keras simple models into Racket several years ago: the built in math library matrix support was fast enough for my needs so the book software may be both pedagogical and useful for smaller problems. EDIT: I have not tried the OpenBLAS Racket bindings here (https://github.com/soegaard/sci) but perhaps the low level tensor and tensor ops book code could be optimized,