PyTorch DataLoader type functionality?

Is there any plan on adding a “Data Loader” capability for Stan? That is, if all of my data cannot fit on a GPU, and I am running a model there, I can batch my data in chunks to feed to the model sequentially.

Furthermore, I can imagine if all of my data won’t fit in ram locally, this could be useful to batch from the hard drive.

Thoughts? Is this on anyone’s roadmap?