Open Songloading opened 2 years ago
@Songloading i have no idea! i'm not really familiar with timm - perhaps you can ask Ross about it?
@lucidrains ok. Any ideas on what else pretrained VIT besides their implementation or pretrained MAE?
Did you solve the issue? @Songloading
Hi There,
I am currently trying to fine-tune an MAE based on pretrained VIT from timm. However, when I do:
I got "AttributeError: 'VisionTransformer' object has no attribute 'pos_embedding'" It seems that timm model is not compatible with the MAE implementation. Can this be easily fixed or I will have to change the internal implementation of MAE?