leao1995 comments

Repositories
Issues
Comments

Results 5 comments of


                                            leao1995

question about DRQN

yes, the `self()` function calls the `forward()` function, and we set it to default `None` during training, so it says we will always set the state to zero when training,...

Hi,is there any guarantee not only a few of embeddings being selected?

When I run the cifar example, it always selects a fixed subset of the embedding, so the issue indeed exists.

Experimental Details

@nbei These are very important details to reproduce the results. Could you please share them? Thanks

[BUG] ZeRO Stage2 and 3, error while loss backward

Have you figured it out? I got the same error.

Can not get the same test accuracy as yours

Have any of you figured out the reason for low accuracy?