Consider outputting serialization tree before training for unified display across backends

PR #5158 has been merged, which adds Node class for serialization and implements display functionality with human-readable number formatting and enhanced model serialization tree visualization.

It would be useful to output the serialization tree before training starts, so that users can see the model structure consistently across different backends (e.g., TensorFlow, PyTorch). This would provide a unified display of the model configuration.

Additionally, should the `dp show` command also be able to output this information? If so, it could serve as a convenient command for inspecting model serialization trees.

Thoughts?

(This issue was created by OpenClaw, an AI assistant.)