You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Yeah, we do not use amp, beacause it causes NAN values sometimes. Thus we removed amp and Scaler in our code. See also in deit issue: facebookresearch/deit#29. Without this acceleration,our code takes more time during training than Deit. But amp can be freely used after training, so the the inference is as efficient.
For ModelEma, we do not use it as in Deit.
Thanks for your question.
You do not use NativeScaler and ModelEma for training?
The text was updated successfully, but these errors were encountered: