Hello, I’m making custom torch.compile backend for our hardware and have a question about torch.fx.Graph.
First, our chip is hard to implement dynamic-shaped tensors (memory map for tensors should be planned before run the model). So my team decided to get maximum size of inputs from users by options keyword argument as a workaround (e.g. options={"max_size": {"x": [32, 32]}}) if the model have dynamic-shaped inputs. But when I printed name of placeholder ops, input names are mangled. Also, there are parameters in placeholder ops.
In this case, how can I find placeholder op that is model’s input? And it is safe to assume inputs are mangled in l_*_ pattern?