2x and 4x Configs and Minor Updates

[mvacaporale]

This includes the most up-to-date 2x and 4x configs as well as

• Small fixes to RigL when gradient_accumulation_steps>1 and prune_fractionn is 0.
• Fix default sparsify_all_embeddings to False - this isn't needed to properly run the 80% sparse model as it's already False in the config. But it's helpful to have there for later.
• The lr range test has been refactored to avoid this bug here https://github.com/pytorch/pytorch/issues/47589
• A small fix in metrics.py when kwargs["lr_scheduler"] is None

[mvacaporale]

There seems to be a caching error with CircleCI. I'm going to close this PR and make a new one to see if that helps.