Can I build PT_v1.12_ZenDNN_v4.0_source_code to obtain a pytorch with both zendnn cpu capability and cuda GPU capability at the same time?

[chencuber]

In Instructions_to_setup_PyTorch_build.txt:
5. Create and activate a conda environment and install the following dependencies
conda install ninja pyyaml cmake cffi typing_extensions future six requests dataclasses astunparse setuptools
conda install cpuonly -c pytorch

why should execute conda install a cpuonly -c pytorch while we actually are building a pytorch from source? Also, can we build a pytorch with both zendnn cpu capability and cuda CPU capability at the same time using "PT_v1.12_ZenDNN_v4.0_source_code"?