I've been trying to resolve broken pytorch installations, and have made several fresh starts by deleting an environment with conda remove -n <env> --all and re-creating it. At some point I noticed that the freshly created environment is not "fresh" at all: when launching python from that environment, I was still able to import the packages from the removed environment of the same name, namely pytorch, which was not installed in the base environment (activate base followed by trying to import pytorch does not work, as expected). I have tried conda create -n <random name> and even the randomly named, empty environment has pytorch inexplicably installed. Running conda clean --all before creating the environment does not prevent this from happening. What is causing this, and how can I prevent this from happening? Continue reading...
