CLIPZeroShotOutput

>>> import lucid
>>> from lucid.models.multimodal.clip._model import CLIPZeroShotOutput
>>> out = CLIPZeroShotOutput(
...     logits=lucid.zeros(2, 10),
...     image_embeds=lucid.zeros(2, 512),
...     text_embeds=lucid.zeros(10, 512),
... )
>>> out.logits.shape
(2, 10)
Ten classes because ten prompts were embedded, not because anything
was trained on ten: the class list is whatever text was handed in.