After watching the launch of Pytorch 2.0, I was very excited about the PrimTorch project, but I had a few questions that were not addressed as I went through the docs and codes.
The current definition of PrimTorch IR (Canonical Aten IR and Prims IR) is incomplete, and one question that bothers me is how PrimTorch IR is designed to ensure completeness of expressiveness? In other words, how is PrimTorch IR designed? I would be grateful for any relevant papers or docs @ mruberry.