Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape `(batch_size, key_value_length)`,

Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape `(batch_size, key_value_l

Preprocess an item to be returned from __getitem__. Override in your child task class, so you have control on what you are returning

Preprocess an item to be returned from __getitem__. Override in your child task class, so you have control on what you are returning

Registers a backward hook on the embedding layer of the model. Used to save the gradients of the embeddings for use in get_gradients

Finds all of the image/text embeddings, and registers a forward hook onto them. When forward() is called, embeddings_list is filled w

Register a forward hook on the embedding layer which scales the embeddings by alpha. Used for one term in the Integrated Gradients su

r""" Extract image & text representations from MMBT. Image representations are the average of image segment in the last layer of MMBT.

Splits punctuation on a piece of text.

Strips accents from a piece of text.

Get values registered in datasets. The dataset builder should save keys of values that is expected to get accessed for model,

Print batch-wise losses/metrics, used in validating a single batch in the training process.

Adds whitespace around any CJK character.

compute the inner product between the difference of parameter and a base parameter and a gradient to compute Eq. (3) in https://arxi

Computes the precision@k for the specified values of k

r"""Aggregates messages from neighbors as :math:`\square_{j \in \mathcal{N}(i)}`. Takes in the output of message computation as first