clip_vit_large_14_zero_shot

→CLIPForZeroShotImageClassification

clip_vit_large_14_zero_shot(pretrained: bool = False, overrides: object = {})

ViT-L/14 CLIP posed as an open-vocabulary classifier.

pretrainedbool= False

Load the OpenAI WIT-400M checkpoint from the Lucid hub.

**overridesobject= {}

Optional CLIPConfig field overrides.

CLIPForZeroShotImageClassification

Scores images against tokenised prompts.

Reference: Radford et al., ICML 2021 (arXiv:2103.00020), Table 11 — the variant the paper's headline zero-shot ImageNet number comes from.

>>> from lucid.models import clip_vit_large_14_zero_shot
>>> model = clip_vit_large_14_zero_shot()
>>> model.config.embed_dim
768