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
First, congratulations on this package! It is amazing, and I think is a true advance for flexibility compared to AutoGL.
One thing that I've noticed in both AutoGL and here is that there's no way to save the model parameters / weights. I've tried to pickle the model, and it works, but when I compare the accuracy of the saved model and the unsaved one, I've got a pretty difference in the accuracy and in all the results. It seems that the weight saving is messing up in some floats.
Ideally, I think the community would benefit from generating a pure PyTorch model from the BoxModel class, so that it can be saved without a problem.
Thanks!
The text was updated successfully, but these errors were encountered:
First, congratulations on this package! It is amazing, and I think is a true advance for flexibility compared to AutoGL.
One thing that I've noticed in both AutoGL and here is that there's no way to save the model parameters / weights. I've tried to pickle the model, and it works, but when I compare the accuracy of the saved model and the unsaved one, I've got a pretty difference in the accuracy and in all the results. It seems that the weight saving is messing up in some floats.
Ideally, I think the community would benefit from generating a pure PyTorch model from the BoxModel class, so that it can be saved without a problem.
Thanks!
The text was updated successfully, but these errors were encountered: