[Training] Cannot export model for inferencing from session created from buffers

[riccardopinosio]

Describe the issue

Currently, if you create a training session with the C api using CreateTrainingSessionFromBuffer() there is no way to export the final model using ExportModelForInferencing(), because the latter complains that no eval path has been provided.

It should be possible I believe to allow for the export of the final model even if the session has been created from buffers.

To reproduce

Just initialize any training session with any model from buffers and try to export the model.

Urgency

No response

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

v2.18.0

PyTorch Version

2.3.1

Execution Provider

Default CPU

Execution Provider Library Version

No response