Hello,
Is there any plan in PyTorch’s roadmap to enable the retrieval of a GraphModule with Torch IR directly in the future? To the best of my knowledge, tracing the model with torch.export.export directly will result in GraphModule with Aten IR. Therefore, I am currently using torch._dynamo.export(model,...,aten_graph = False)(inputs) to trace GraphModule in which basic torch.nn.Modules (e.g., Linear layer, matmul) remain as leaf nodes, as I am trying graph optimization on Torch IR level. I’m wondering if using torch._dynamo.export will remain the recommended approach for my purpose of interest moving forward, or if there would be any upcoming changes that might make it easier or more efficient to obtain the graph module with Torch IR. The only concern I have is that the underbar in “_dynamo” might implicate that it can be discarded in future torch release 