(Scythian? cython…) Rust had packaging, great tooling for testing and docs, great FFI for Python, AND could be compiled to wasm. And a much better story for refactoring and maintaining code in the long term, which was starting to become a strain in the Python parts of sourmash

We’re looking for a volunteer reviewer with Cython expertise and an interest in network diffusion processes to help review the CyNetDiff Python package. 🫶 New to reviewing? We offer mentorship! 💻 Learn more about CyNetDiff: https://buff.ly/3XI9jaAbuff.ly/4dxYpua

GitHub - eliotwrobson/CyNetDiff: A performance-focused library implementing algorithms for simulating network diffusion processes, written in Cython.

A performance-focused library implementing algorithms for simulating network diffusion processes, written in Cython. - eliotwrobson/CyNetDiff

I’ve had an unfortunate trend lately of presenting on the last day at conferences 😅 That being said, just finished presenting “Sleuthing in Cython” on memory profiling adventures at #pyconjp2024pretalx.com/media/pyconj...#cython#oss#python

Cython faz milagre

“Cython, pybind11, cffi – which tool should you choose?” blog.behnel.de/posts/cython...#opensource#Python#Programming#cpp

（HPC的な意味で）行列積と戯れてたらこんな時間に。 JITコンパイラがブラックボックスにnative化+O3最適化を走らせるからLLVMのloop展開やらレジスタの乗せ方の挙動がいちいち謎で格闘してた（C++/gccと違う）（てかなんでこんな黒魔術をPythonで書いてんだよ…でもおかげでカリカリ実装をCythonやpybind無しで錬成できるようになった）。 CPUでこんな泥団子こねくり回すよりCUDAでflash attentionとか実装するほうがたぶんラクで楽しいのでは。 まあ一応目標値までは速くなったんですけど。

Rust is a huge improvement over C to be sure! But I personally find it a very difficult language, so it’s less likely that I’d be able to execute something in Rust myself. And things like Cython never really got excellent, so I get why we moved on, but I can’t help but wish it had worked out.

対局するところだけC++に移植してCythonでラップするとだいぶ速そう あとは盤面管理とかをビットボードにするとか？(Pythonでは効果ないと思う

Python (Cython)でsummaly proxy書いた​:thinkhappy:​ (この投稿に付いてるやつみたいなのを取得するためのシステム) <github.com/AmaseCocoa/summaly-py>

Installation requires FFMPEG and Cython, and usage involves running `diarize.py` with the audio file name. There is an experimental parallel version `diarize_parallel.py` for systems with sufficient VRAM. Command line options and known limitations are listed, with future improvements planned. (2/3)