It is possible that this is due to a bug or awkward specification in the Zero GPU space.
It seems that aqlm is supposed to reference CUDA_HOME, but there is no CUDA Toolkit installed in the Zero GPU space to begin with.
I’ll try to fix it, but I’m not sure if it will work or not. loading the model into CUDA and packing the tensor works fine, so I have a feeling it will work if the libraries can handle it.