Yeah that’s annoying but it’s a short-term problem. Python just recently cleaned up some long-standing issues that broke backwards compatibility in packaging (for certain things). Most public modules that broke made trivial changes to fix the problems (once they learned about them) and life went on.
However, for some fucking reason a whole bunch of dependencies related to AI are dragging their feet and taking forever to fix their shit. Insisting that everyone “just use Python 3.10” and it drives me nuts too.
This problem started to become a real thing almost two years ago (so they had plenty of warning and time to fix things) and yet here we are with still a handful of core dependencies that won’t install for things like Stable Diffusion, Flux, and various LLM stuff because they’re dragging their feet.
I blame corporate culture: Enterprises hate upgrading their shit and they’re as slow as glaciers sometimes. There’s probably tooling at Nvidia, for example, that needs a ton of work for Torch to work with new versions of Python and since all their documentation already was written for running on Python 3.10 (and Ubuntu 22.04 LTS) they’ve created a lot of work for themselves.
Any day now they’ll finally finish fixing all these little dependencies and then we’ll have another two years of ease before the problem rises again with Python 3.14 and it’s massive GIL-free improvements that require big changes in code to actually take advantage of them.
Yeah that’s annoying but it’s a short-term problem. Python just recently cleaned up some long-standing issues that broke backwards compatibility in packaging (for certain things). Most public modules that broke made trivial changes to fix the problems (once they learned about them) and life went on.
However, for some fucking reason a whole bunch of dependencies related to AI are dragging their feet and taking forever to fix their shit. Insisting that everyone “just use Python 3.10” and it drives me nuts too.
This problem started to become a real thing almost two years ago (so they had plenty of warning and time to fix things) and yet here we are with still a handful of core dependencies that won’t install for things like Stable Diffusion, Flux, and various LLM stuff because they’re dragging their feet.
I blame corporate culture: Enterprises hate upgrading their shit and they’re as slow as glaciers sometimes. There’s probably tooling at Nvidia, for example, that needs a ton of work for Torch to work with new versions of Python and since all their documentation already was written for running on Python 3.10 (and Ubuntu 22.04 LTS) they’ve created a lot of work for themselves.
Any day now they’ll finally finish fixing all these little dependencies and then we’ll have another two years of ease before the problem rises again with Python 3.14 and it’s massive GIL-free improvements that require big changes in code to actually take advantage of them.