I saw this pr Remove dynamo+nvfuser by IvanYashchuk · Pull Request #105789 · pytorch/pytorch · GitHub which removed nvfuser from pytorch dynamo. Can I still use nvfuser as the backend in torch.compile in the future? Has nvfuser been completely replaced by triton in the Torch graph compilation system?
the last discussion we had around nvFuser was that nvFuser maintainers thought it would benefit out-of-core development for a little bit, also try take it in a few different directions – possibly some pytorch-agnostic ones. I was supportive of them exploring that direction.