GPU memory

Hello and thanks for this very nice job.
I want to ask if the code could be refactored and use Distributed Data Parallel instead of Data Parallel. Or if you have a tip for me about the implementation, so i can do it.

thanks

The issue names no files, tests, or entry points. Start by locating the current Data Parallel implementation and the training entry point, then clarify the supported Distributed Data Parallel scope with maintainers. Done would mean an agreed, tested replacement rather than implementation advice alone.

Written by the indexing model from the issue text.