Args: contextualized_embeddings: Hidden states for contextual layer. batch: A dict of a batch. mask: Mas

Compute a weighted average of the ``tensors``. The input tensors an be any shape with at least two dimensions, but must all be the same shape

Input is expected to be of size [bsz x seqlen]. Args: inputs: Tensor: Returns:

Args: x: batch_size x max_len x d_model mask: batch_size x max_len Returns:

Args: x: batch_size x max_len x hidden_size mask: batch_size x max_len, 为0的地方为pad Returns: batch_size

Args: x: batch_size x max_len mask: batch_size x max_len. 有value的地方为1 x: Tensor: mask: Tensor:

Compute the conditional log likelihood of a sequence of tags given emission scores. Args: emissions (`~torch.Tensor`): Emission s

Creates an optimizer from its config with WarmUp custom object. Args: config: Returns:

Args: config: kwargs: config: dict: Returns:

Build an object from config. Args: config: A ``dict`` holding parameters for its constructor. It has to contain a `classpath` key,

Parameters ---------- meta kwargs Returns ------- KerasComponent

r''' * Takes in two lists of matrics `state_S` and `state_T`. Each list contains two matrices of the shape (*batch_size*, *length*, *hidden_size*)

Create a sequence of x for tokenization task. Each x is an atomic subtoken that will be tagged with BMES or BI tags. Args: sample: D

Returns the device of the tensor. Args: tensor: torch.Tensor: Returns:

Modified from https://github.com/huggingface/transformers/blob/7b75aa9fa55bee577e2c7403301ed31103125a35/src/transformers/trainer.py#L232 Setu