Evaluate the model on a given split. Args: split_name (str): name of the split to evaluate on. cur_epoch (in

Converts all `BatchNorm2d` and `SyncBatchNorm` layers of provided module into `FrozenBatchNorm2d`. If `module` is itself an instance of eithe

Makes broadcastable attention and causal masks so that future and masked tokens are ignored. Arguments: attention_mask (

Create the directory if it does not exist.

Compresses the memory bank if the current memory bank length is greater than the threshold. Compression_size is the number of frames that are

r"""Register an item to registry with key 'name' Args: name: Key with which the item will be registered. Usage::

Organizes datasets by split. Args: datasets: dict of torch.utils.data.Dataset objects by name. Returns: Dict of dataset

Args: samples (dict): A dictionary containing the following keys: - image (torch.Tensor): A tensor of shape (batc

Saving the results in the format required for leaderboard evaluation. [TODO] add support for multi-choice.

An inner training loop compatible with both epoch-based and iter-based training. When using epoch-based, training stops after one ep

Performs all_gather operation on the provided tensors. Graph remains connected for backward grad computation.

Returns list of utf-8 byte and a corresponding list of unicode strings. The reversible bpe codes work on unicode strings. This means you

Concatenates multiple datasets into a single dataset. It supports may-style datasets and DataPipeline from WebDataset. Currently, does not s

Convert applicable model parameters to fp16

Convert applicable model parameters to fp16

Create dataloaders for training and validation.

Args: clip (torch.tensor): Video clip to be cropped. Size is (C, T, H, W)

Get question ids that satisfy given filter conditions. default skips that filter :param imgIds (int array) : get question ids f

Return set of symbol pairs in a word. Word is represented as tuple of symbols (symbols being variable-length strings).

Load from a finetuned checkpoint. This should expect no mismatch in the model keys and the checkpoint keys.

Example >>> dataset = load_dataset("coco_caption", cfg=None) >>> splits = dataset.keys() >>> print([len(dataset[split]) for split in

A property to get the DDP-wrapped model on the device.

Rotates half the hidden dims of the input.

Warmup the learning rate