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 we had the pre-trained weights included in the package, but that made it really heavy.
Next, we moved those to Google Drive, and these are download using gdown when a model is initialized.
I think Google Drive is a good patch for now but it's not the appropiate way. Maybe we should use Zenodo or some similar repository to safely store the weights online, retireve them from there, and that would allow a better tracking of different versions.
Maybe then we should allow load_model() to have a version argument so that users can initialize the different versions online, and always default to the last one.
The text was updated successfully, but these errors were encountered:
Alright:
gdown
when a model is initialized.I think Google Drive is a good patch for now but it's not the appropiate way. Maybe we should use Zenodo or some similar repository to safely store the weights online, retireve them from there, and that would allow a better tracking of different versions.
Maybe then we should allow
load_model()
to have aversion
argument so that users can initialize the different versions online, and always default to the last one.The text was updated successfully, but these errors were encountered: