A node type in export IR graph

torch.compile FX (Functional Transformations)
1

I’m working on graph transformation with export IR graph in ExportedProgram after torch.export.export.

When I implemented a pass for the transformation, I had one question about a node type. For example, this constant propagation pass in executorch codes replaces redundant torch.fx.Node with torch.Tensor.

But, as a spec says, all the nodes are torch.fx.Node.

A Node represents a particular computation or operation and is represented in Python using the torch.fx.Node class.

Is it okay to use torch.Tensor, too? It wouldn’t violate the spec?

2

Ah, I missed some points. The pass internally creates placeholder fx node instead.