-
Notifications
You must be signed in to change notification settings - Fork 3.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Accuracy metric return tuple of length num_gpus on ddp in 0.9.1rc4 #3676
Comments
Thanks for the report. Apologies but I'm afraid I don't understand how this relates to our accuracy metric. |
This is my code:
0.9.1rc4:
In 0.9.1rc4, |
yes it does. |
I write a example to reproduce:
Here is error:
|
ok, can reproduce. not sure yet why the two behave differently |
The issue is that functional accuracy directly calculates it and modular accuracy also does some syncing. But it definitely shouldn't be like this. I'll have a look at it. |
This has to do with how aggregation currently is implemented in the class metrics is implemented. |
code:
In 0.9.1rc3, vid acc is a tensor, but in rc4, it changes to a tuple. I want to use
-vid_acc
as monitor, and I think it should be a tensor.Using rc4, in macos's cpu mode, it's a tensor. But in linux ddp mode, it's a tuple of length
num_gpus
.The text was updated successfully, but these errors were encountered: