TMVA_SOFIE_ONNX.C tutorial fails with modern PyTorch due to tensor name mismatch

[Neeraj-x0]

Check duplicate issues.

• Checked for duplicates

Description

The tutorial TMVA_SOFIE_ONNX.C hardcodes a lookup for a tensor named "16weight"
(line 39), but the companion script TMVA_SOFIE_ONNX.py generates an ONNX model
where tensors are named "0weight", "2weight", "4weight" etc. (PyTorch default
Sequential module naming). This causes a runtime_error when running the C++ tutorial.

Expected behavior: The tutorial should run end-to-end without manually providing
a pre-existing ONNX file.

Workaround: Copy tmva/sofie/test/input_models/Linear_16.onnx into the build
tutorials directory manually.

Reproducer

1. Run the Python script to generate the ONNX model:
   cd $ROOTSYS/tutorials/machine_learning
   python3 TMVA_SOFIE_ONNX.py

2. Run the C++ tutorial:
   root -l TMVA_SOFIE_ONNX.C

Error:
TMVA SOFIE tensor [16weight] for which the shape is requested is not found
terminate called after throwing an instance of 'std::runtime_error'

The Python script generates tensors named "0weight", "2weight", "4weight"
but the C++ tutorial looks up "16weight" which does not exist in the model.

ROOT version

6.39.01

Installation method

Built from source

Operating system

Ubuntu 24.04

Additional context

PyTorch version: 2.10.0+cu128
The correct Linear_16.onnx already exists in tmva/sofie/test/input_models/
and works correctly when copied manually to the build directory. The tutorial
and the Python script appear to be out of sync.