Description
As of 525fdc0 torch.utils.cpp_extension.load() does not support building SYCL kernels. Please, consider adding such support. It's used (load API), for example, by https://github.com/dvrogozh/optimum-quanto. See invocation here.
Note: I am willing to help with implementation and will provide a PR with the proposal shortly. Update: see #132945
Update 8/16: other APIs, load_inline and class API is not supported as well and also need to be implemented.