Skip to main content


I’m stuck trying to install the #Python `torch` module with #CUDA support on an old iMac with #Debian reddit.com/r/LocalLLaMA/commen…

Anyone have experience here?

#AI #Llama

Unknown parent

mastodon - Link to source
Curtis "Ovid" Poe (he/him)
@julia Darn it. I'm told Triton might be an option?
in reply to Curtis "Ovid" Poe (he/him)

I ran into an issue with this recently when i was making docker containers that included pytorch as a dependency. I had to install pytorch with the CPU only platform before the python module and its requirements file (with pytorch) ran.

pytorch.org/get-started/locall…

in reply to Adriaan

@adriaan yep, and usually tell people at work to use conda to install all of that

"It just works" usually, but yeah hopefully you don't need an Nvidia GPU otherwise for what you are doing. As others have noted

in reply to Curtis "Ovid" Poe (he/him)

@adriaan it's another virtual environment but like over powered

I found it easier to install many projects that ai/ML folks commonly use from conda than pip

I also support hundreds of folks and hundreds of machines so they don't have root and conda sidesteps a lot of that headache. Mostly for me actually.