data

VJEPA2Output

extendsModelOutput

VJEPA2Output(tokens: Tensor | None = None, prediction: Tensor | None = None, target: Tensor | None = None, loss: Tensor | None = None, context_indices: Tensor | None = None, target_indices: Tensor | None = None)

source edit

Output of a representation pass or masked prediction step.

tokens is populated for a plain representation pass. When both mask index matrices are supplied, prediction, target and loss hold the predictor result and the representation-space training objective.

Used by 1

lucid.models.vision.vjepa2

Constructors

dunder

init

→None

__init__(tokens: Tensor | None = None, prediction: Tensor | None = None, target: Tensor | None = None, loss: Tensor | None = None, context_indices: Tensor | None = None, target_indices: Tensor | None = None)

source edit