Torch.fx may be have a bug in torch2.6

1

When modifying the parameters of the GraphModule node, the modified network returns a value of None.

eg:

for node in model.graph.nodes:
    node.args = new_args

torch.fx.symbolic(model, model.graph)

# when forward this model, will return None, in other pytorch version have not this phenomenon.
2

2.6 is quite old by now so did you see the same issue in the latest stable or nightly release?