How does cosine lr scheduler work when the prodigy optimizer is used ?

The prodigy (or Dadapt) calculate lr itself. I know the lr rate argument is actually the ratio on that calculated lr. Does it also work that way if I use cosine_with_restarts? Will the ratio change as the way when a AdamW is used?

Start by locating the optimizer and scheduler entry points for Prodigy/DAdapt and cosine_with_restarts, then compare their learning-rate handling with AdamW. Done means a clear, repository-backed explanation of whether the ratio changes under cosine restarts.

Written by the indexing model from the issue text.