Calculate lr decay rate for different ViT blocks. Args: name (string): parameter name. lr_decay_rate (float): base lr decay r

Inference with flip augmentation

Inference interface for the model Args: x (torch.Tensor): input tensor of shape (b, c, h, w) pad_input (bool,

Inference interface for the model with horizontal flip augmentation Horizontal flip augmentation improves the accuracy of the model b

Returns: True if the current process is the main one.

log(nCk) using stirling approximation

Pairwise nearest neighbors for L2-normalized vectors. Uses Torch rather than Faiss to remain on GPU.

Parse a list of values for the key if the value is a string. The values are separated by a comma. Modifies the config in place.

Generate positional encodings Args: sequence_length (int): Sequence length embedding_dim (int): Embedding dimension

cycles through iterables but sample wise first yield first sample from first iterable then first sample from second iterable and so on th

Splits the arguments that are combined with '__' into multiple arguments. Combined arguments should have equal number of keys and values.

Tag the last checkpoint. Args: last_filename_basename (str): the basename of the last filename.

Args: pic (PIL Image, numpy.ndarray or torch.tensor): Image to be converted to tensor. Returns: Tensor: Conve