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
I was wondering if you had considered sharing the 30th, 60th, and 90th Resnet 50 checkpoints you are using in the notebook. Without them, like karasepid in another issue/thread, I'm sometimes confused about the shape of some of the variables/fields used. If the checkpoints were made available, then it would be straightforward for us to just step through the notebook code and print out those shapes.
Could the checkpoints perhaps be shared from a Google drive, or whatever online storage works best for you?
Thanks for considering this request!
Best regards, -- Phil
The text was updated successfully, but these errors were encountered:
I'm really sorry philferriere@. Checkpoint release requires Google's legal approval and I don't think I will be able to do that. To validate the shapes, I think you can simply comment out weight loading and use 3 random inits.
Using 3 random inits is exactly what I've been doing to familiarize myself with the notebook code. The results are (obviously) meaningless but I am at least able to confirm tensor shapes.
Thank you for sharing the rational for not distributing the checkpoints though.
Hi @frederick0329 !
I was wondering if you had considered sharing the 30th, 60th, and 90th Resnet 50 checkpoints you are using in the notebook. Without them, like karasepid in another issue/thread, I'm sometimes confused about the shape of some of the variables/fields used. If the checkpoints were made available, then it would be straightforward for us to just step through the notebook code and print out those shapes.
Could the checkpoints perhaps be shared from a Google drive, or whatever online storage works best for you?
Thanks for considering this request!
Best regards, -- Phil
The text was updated successfully, but these errors were encountered: