Does this compile a hipified Cuda-kernel meant for Pytorch for CPU?

[rmast]

I want to try this Cuda-dependent project on CPU: https://github.com/retsuh-bqw/SRFormer-Text-Det
However, I don't have a CUDA-GPU with sufficient memory.
The project contains a single Cuda-kernel without a valid CPU translation:
https://github.com/retsuh-bqw/SRFormer-Text-Det/blob/main/adet/layers/csrc/DeformAttn/ms_deform_attn_cuda.cu

The CPU build breaks on this 'Not implemented' error:

https://github.com/retsuh-bqw/SRFormer-Text-Det/blob/main/adet/layers/csrc/DeformAttn/ms_deform_attn_cpu.cpp

Can I somehow HIPify the CUDA-kernel to HIP to make Pytorch use the hipified version on CPU?

Should I use Pytorch HIP for that purpose or will regular pytorch do?