for the two parts of the model you might call the "encoder" and the "classifier". However, the latter is called "overall" in flags/doc names, which is misleading since those parameters are not applied overall at all...but rather just to the classifier layers. Therefore we should do a search-and-replace for overall_ and replace it with classifier_.
Right now we have separate
for the two parts of the model you might call the "encoder" and the "classifier". However, the latter is called "overall" in flags/doc names, which is misleading since those parameters are not applied overall at all...but rather just to the classifier layers. Therefore we should do a search-and-replace for
overall_and replace it withclassifier_.