Thank you! + Some benchmarks

[andrewdalpino]

Dear Zephir team,

I just wanted to say thank you for all your work. Zephir has played a key role in our development of the Tensor extension which is an optional package that you can install to accelerate machine learning algorithms in Rubix ML and hopefully more libraries in the future. For us, Zephir really shines when the API code is written mostly in Zephir and the important underlying implementations are written in C and then connected using Custom Optimizers. We get the ease of use and quick iteration that comes with Zephir and the ability to optimize the code down to bare metal with C.

I also wanted to share with you some recent benchmarks that we are very happy with. By picking off some low hanging fruit using BLAS routines powered by OpenBLAS we were able to speed up training on our MNIST example project by 12X - reducing the training time from hours to just minutes.

We're hoping to unlock a whole other set of low-hanging fruit once we figure out how to store C data alongside the objects like they describe here.

So thank you again and keep up the great work!

Andrew