Dirk van den Bekerom

I'd be interested in this as well, I don't have time at the moment but perhaps in some future..

Parsing with custom c code (including SIMD intrinsics) should be extremely fast. Also: we could process one dataset while downloading the next (maybe we already do this?) https://docs.python.org/3/library/asyncio.html https://pypi.org/project/aiohttp/

I have worked on a C++/SIMD implementation which was very fast, I will post more details over the weekend

Here's a simd C++ code that reads super fast; download became the bottleneck after that: ``` // loadHitranSIMD.cpp : This file contains the 'main' function. Program execution begins and ends...

Very nice graph! The size of a mini-batch is larger than L+M-1 though. In `scipy` this is called the `block_size`, it is calculated by the Labert w function and is...

For my understanding, is this basically what SPEARS does? if not, how is it different?

It seems at least for the C++ part, [`cibuildwheels` is how some of the larger packages do it](https://github.com/pypa/cibuildwheel#working-examples) For the GPU part, most packages I know of require the CUDA...

> Btw the deploy step of Radis (activated on pushes to the master branch) already builds a wheel That's interesting.. It seems it makes [one wheel](https://pypi.org/project/radis/#files) for all platforms and...

We will have to decide: what is the lowest python version we would want to support?

Indeed vulkan does not use cython, i think you can remove it I find the easiest way to implement compiled code is through separate dlls or precompiled python modules using...