Skip to content
This repository has been archived by the owner on Mar 19, 2024. It is now read-only.

EMA does not work on fp16 and does not copy weights? #568

Open
YannDubs opened this issue Sep 14, 2022 · 0 comments
Open

EMA does not work on fp16 and does not copy weights? #568

YannDubs opened this issue Sep 14, 2022 · 0 comments

Comments

@YannDubs
Copy link

Hi,

Looking at the implementation of ModelEmaV2, it seems that compared to timm the model only works on fp32 parameters? (see this line )
Does it mean that it will not work if I use AMP ?

Furthermore, another difference with timm is that the ema_model is not copied (in timm copying is done here ). I am probably missing where the model is copied, can you point it to me please? (if the model is not copied then EMA simply corresponds to momentum)

@YannDubs YannDubs changed the title EMA does not copy weights and does not work on fp16? EMA does not work on fp16 and does not copy weights? Sep 14, 2022
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant