Choose parallelism and size your training GPUs.
Recent revisions are kept. Restoring one creates a new revision.
