I hate venv so much it's unreal. Each venv takes 10 gb of space with PyTorch and other shit. Isn't there a way to do this without wasting hundreds of gigabytes of space? I'm moronic if you need to know.
I hate venv so much it's unreal. Each venv takes 10 gb of space with PyTorch and other shit. Isn't there a way to do this without wasting hundreds of gigabytes of space? I'm moronic if you need to know.
install pytorch system wide, then
virtualenv --system-site-packages newvenv
isn't pipx a better way to sandbox and manage python packages ?
pipx upgrade-all
there are like 40 solutions to this problem and none of them are very good
pipx is exclusively for executables, not libraries, that's the entire point
>muh space
Black person you're gonna use it to run 30GB+ LLMs anyway.
Yes and I don't want to put PyTorch on my NAS server because it'll take ages to load.
>passgay
>moronic Black personhomosexual
checks out
Bad solution, there are other Python packages installed system wide that's uncompatible with specific programs. I only need PyTorch and some other huge libraries.
I looked into it but isn't pipx more focused on executables?
Now that's interesting. I'll look into that. Also, what's Conda's deal? People seem to use it over the builtin venv. Does it do something similar too?
I'm poor.
check out https://github.com/astral-sh/uv
uv pip uses a global cache, but it's more for performance than storage use
There's a million ways to manage environments and installations
I unironically spin a new VM every time I need to work on a different Python project
I hate Python so much bros
Are you poor? Just get a bigger ssd.
>python
>batteries included
heh
>Isn't there a way to do this without wasting hundreds of gigabytes of space
Yeah, it's called delete Python and start writing your code in C. And stop importing libraries unless you can't rewrite the functionality of that library in a single day.
>writing your code in C
Reinventing the wheel intensifies
Reinventing the wheel is better than whatever the frick npm/cargo/pipenv is
the cuda libs are C libraries anon
Oh I recognize that fizzbuzzer.
What's up bro, you're still gonna write a book on machine learning in C because you figured out how to write a GEMM?
>pip install package --break-system-packages
>reinstalling requirements.txt every time instead of switching environments
i kneel...
no hehe
just be thankful that it works (more or less)
im sure you COULD make it smaller if you tried but it's easier to just buy more storage
If you don't plan on updating the different venvs often, just use hard links to make all common files to be shared.
https://www.jdupes.com/
Ironically runtimes like in Flatpak partially solve this issue.
>filtered by python
ngmi
Imagine needing to care about where your dependencies are stored