Training a smaller model

[ShivanshuPurohit]

To reduce the model size for a single GPU, is there anything more to do after reducing the number of residual blocks in vae.py?

[rewonc]

Sure, you could reduce the width parameter (in addition to reducing the number of layers in the decoder or encoder)

[ShivanshuPurohit]

Any other tips to train on single V100?

[rewonc]

You can just train on a single V100 with the hyperparams as given. You'll probably have to decrease the learning rate, because the batch size will be lower.

…

On Fri, Dec 18, 2020 at 7:15 AM Shivanshu Purohit ***@***.***> wrote: Any other tips to train on single V100? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAYEDVA6EOF2FDEJX3LUY6LSVNW2FANCNFSM4U4CQZSQ> .