Azure ML define transformer path from a data asset/datastore within the config.cfg file

[EY4L]

I am attempting to train a transformer based model in Azure ML.

Currently each time a job is run the roberta-base model is being downloaded on each job run.

I wanted to find out if anyone has an approach to how to define paths in the config file so that a training job can either read the transformer weights from a datastore, a data asset or from the model registry.

Alternatively any other solutions within a cloud based training system would be appreciated if shared. Or what general approach could be taken to solve this issue.

Thank you very much.