:param text_embed_l1: bsz_text, hidden :param video_embed_l1: bsz_video * n_clips, hidden :return: bsz_text, b

:param text_embed_l1: bsz_text, hidden :param video_embed_l1: bsz_video * n_clips, hidden :return: bsz_text, b

:param text_embed_l1: bsz_text, hidden :param video_embed_l1: bsz_video * n_clips, hidden :return: bsz_text, b

Return the latest inserted value.

Get the block args and global params for a given model name. Args: model_name (str): Model's name. override_params (dict): A dict

display each optimizer's learning rate for logging purpose during training

Get index of padding <pad> token in vocabulary. Returns: int: index of the padding token.

Args: # sim_matrix_semantic: [bt, bv] sequence_output: sequence tuple, (sentence-level, word-level feats)

Args: seq_len: int, the sampled indices will be in the range [0, seq_len-1] num_samples: sample size devi

Args: image: PIL Image or torch.tensor max_size: Returns: Note the height/width order difference

Get the latest value of each metric. Returns: dict: a dict which key is the metric name and the value is the latest valu

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

Override this in your child class to extract custom text Default for VQA. Make sure to return a list of all possible text

function to locate absolute paths of user model related resources **Note**: for online serving, user resource files must be accessed by this

Get size of the vocabulary. Returns: int: size of the vocabulary.

Get vocab size of the answer vocabulary. Can also include soft copy dynamic answer space size. Returns: int: size of the

Obtain height and width from x. Args: x (int, tuple or list): Data size. Returns: size: A tuple or list (H,W).

r"""Returns batch-wise graph-level-outputs by averaging node features across the node dimension, so that for a single graph :math:`\mathcal{G}

r""" during adversarial training, the model needs to be in eval mode to avoid randomness such as drop-out

Provide default implementation based on preprocess_cfg and user can reimplement it. if nothing to do, then return lambda x: x

Initializes and prunes weights if needed.