:param sim_matrix: #text(bsz) x #video(bsz*n_pair) :return:

:param sim_matrix: #text(bsz) x #video(bsz*n_pair) :return:

:param sim_matrix: #text(bsz) x #video(bsz*n_pair) :return:

Calculate the 3D rope index based on image and video's temporal, height and width in LLM. Explanation: Each embedding se

:return: bsz_text, bsz_video, num_clips: clip-level score for training bsz_text, num_clips: clip-level score for val/

Whether the configuration is frozen.

Print a list of json dictionary in a file The file then contains information such as {xxx:yyy} {zzz:qqq} Parameters:

FSDP module .apply will use _unshard_params_recurse which will sync params across ranks. using this function when apply fn is unnecessary to sync

Rotates half the hidden dims of the input.

Returns the closest integer to 'number' that is divisible by 'factor'.

Add salt noise, pepper noise, or, s&p noise to img. @param img: ndarray Input image data on range[0,255]. @param amount : floa

:param encoder_embeds: (b, sq_length, hidden) :param position_range: :return:

:param encoder_embeds: (b, sq_length, hidden) :param position_range: :return:

:param encoder_embeds: (b, sq_length, hidden) :param position_range: :return:

:param encoder_embeds: (b, sq_length, hidden) :param position_range: :return:

Override the default train() to allow freezing image encoder, text encoder, and classifier :return:

Update the config in an inplace way. Usage: .. code-block:: python config: Configuration = ... upd

Use this if there is some specific configuration required by model which must be inferred at runtime.

Runs basic whitespace cleaning and splitting on a piece of text.

Args: data: image bytes json_obj: one json object include multimodal info Returns: sample(antmm