I am sitting on a linux machine with a working pytorch installation and cuda drivers. If I simply make an export of my current conda environment, it will not use the rocm pip wheels, but will instead include a lot of cuda libraries. What is your recommended recipe for making a rocm-compatible conda environment, starting from a working cuda environment?
I hope the question makes sense.
edit: I see that you give example code for how the py39_pytorch_rocm.yml file was created. I suppose that answers my question :)
I am sitting on a linux machine with a working pytorch installation and cuda drivers. If I simply make an export of my current conda environment, it will not use the rocm pip wheels, but will instead include a lot of cuda libraries. What is your recommended recipe for making a rocm-compatible conda environment, starting from a working cuda environment?
I hope the question makes sense.
edit: I see that you give example code for how the py39_pytorch_rocm.yml file was created. I suppose that answers my question :)