Pretrained weights for lucid.models.clip_vit_large_14.
The paper's best variant, and the only one whose text tower is widened along with the image tower — 768 wide with 12 heads against the base models' 512 and 8, at the same 12 layers.
Examples
>>> from lucid.models.multimodal.clip._weights import CLIPViTLarge14Weights
>>> list(CLIPViTLarge14Weights.__members__)
['OPENAI_WIT400M', 'DEFAULT']
DEFAULT is an alias rather than a fourth entry, so a bare
pretrained=True and the tag it resolves to cannot drift apart.
>>> CLIPViTLarge14Weights.DEFAULT is CLIPViTLarge14Weights.OPENAI_WIT400M
True
>>> CLIPViTLarge14Weights.OPENAI_WIT400M.num_classes
0