How to uninstall/remove PyTorch when built and installed from source?

1

I followed this guide to install the latest PyTorch with Cuda support on my system.

and importing torch in a terminal I can see the following:

torch.__version__
'2.4.0x0+gitf14cdc5'

But the problem now is that that installation overwrites all the other different pytorch versions in my other environments. Activating other conda environments show me exactly the same pytorch version.

Now, how do I get rid of the build and installed version of pytorch?

Thanks

EDIT: I read in many other threads, that when you install PyTorch using conda then it comes with it own cuda environment and so on. But I do not understand, why the installation from source did overwrite all my pytorchs on my system. If I create a new environment and install pytorch using conda, then it recognize always and everywhere the installed version from source.

2

Build PyTorch in an own environment instead of the base env.

3

This is exactly what I did, because I didn’t want to infest my server. I created a new conda environment then I built the conda following the tutorial.

I even looked in stackoverflow for similar problems. Usually there should be a
$ make uninstall
command in torch but it seems there is not.

4

I have 35 virtual environments in conda on my current system with stable, nightly, and source-built binaries spanning from torch==0.4.1, 1.12.2+cu113, to the latest nightlies using CUDA 12.4 including source builds using my locally installed CUDA toolkit and have never seen that any libraries “leak” from one virtual environment to another one (not speaking of the base env which is entirely empty on my system).

make uninstall does not exist, but pip uninstall torch -y will do it.