Basically, I want to know how many forward passes are saved when using the option of assistant_model.
It would also be ideal to know how many tokens from the assistant_model were accepted.
Is there anyway to get these kind of diagnostics when using assistant_model?