Number of hypotheses in the list.

Beam search Args: model_inputs (dict): input ids. beam_size (int, optional, defaults to 3): beam size of beam

Top-k and top-p sampling. Args: model_inputs (dict): input ids generate_length (int, optional, defaults to 10

Beam search Args: model_inputs (dict): input ids. beam_size (int, optional, defaults to 3): beam size of beam

Compact the dataset, removes blocks which the files were deleted. **Note** This may affect the existing dataset state dict. Args: pa

Decode ids into a string.

Args: x (:obj:`torch.Tensor` of shape ``(batch_size, seq_len, dim_norm)``): Input tensor that need to be normalized. Retu

Args: hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Input of self-attention block. It can be the

Args: hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Hidden states before feed forward layer.

Args: self_hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Input of transformer block(self-attenti

Args: x (:obj:`torch.Tensor` of shape ``(batch, seq_len, dim_in)``): The input of linear layer Returns: :obj:

Args: x (:obj:`torch.Tensor` of shape ``(..., dim)``): Inputs. x_pos (:obj:`torch.Tensor` of shape ``(...)``): Positi

This model inherits from bmt.DistributedModule. Args: hidden_q (:obj:`torch.Tensor` of shape ``(batch, len_q, dim_model)``): Indic

Args: hidden-states (:obj:`torch.Tensor` of shape ``(batch, seq_enc, dim_model)``): Input of encoder, might be the embedding of a

Args: ids (:obj:`torch.Tensor` of shape ``(batch_size, seq_len)``): Indices of input sequence tokens. Return:

Args: ids (:obj:`torch.Tensor` of shape ``(batch_size, seq_len)``): Indices of input sequence tokens. ids (:obj:`torc

This model inherits from bmt.DistributedModule. Transform an input tensor from one feature space to another via a nonlinear operation

Args: x (:obj:`torch.Tensor` of shape ``(batch, seq_len, dim_in)``): The input of feed-forward module. Return:

Args: x (:obj:`torch.Tensor` of shape ``(batch_size, seq_len, dim_norm)``): Input tensor that need to be normalized. Retu

Args: hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Input of self-attention block. It can be the

Args: hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Hidden states before feed forward layer.

Args: self_hidden_states (:obj:`torch.Tensor` of shape ``(batch, seq_self, dim_model)``): Input of transformer block(self-attenti

Args: x (:obj:`torch.Tensor` of shape ``(batch, seq_len, dim_in)``): The input of linear layer Returns: :obj:

Args: x (:obj:`torch.Tensor` of shape ``(..., dim)``): Inputs. x_pos (:obj:`torch.Tensor` of shape ``(...)``): Positi

Args: hidden_q (:obj:`torch.Tensor` of shape ``(batch, len_q, dim_model)``): Indices of input sequence tokens. It will be embedde

Args: hidden-states (:obj:`torch.Tensor` of shape ``(batch, seq_enc, dim_model)``): Input of encoder, might be the embedding of a

Args: ids (:obj:`torch.Tensor` of shape ``(batch_size, seq_len)``): Indices of input sequence tokens. Return:

Args: ids (:obj:`torch.Tensor` of shape ``(batch_size, seq_len)``): Indices of input sequence tokens. ids (:obj:`torc

Transform an input tensor from one feature space to another via a nonlinear operation Args: x (:obj:`torch.Tensor` of shape ``(ba

Args: x (:obj:`torch.Tensor` of shape ``(batch, seq_len, dim_in)``): The input of feed-forward module. Return:

Load dataset state. Args: state (dict): dataset state dict. strict (bool): If `strict` is True, world size needs to b

Mask one file in dataset. Blocks in masked datasets won't be read later. Args: path (str): path to dataset dbname (str): file nam

Projection based on embedding's weight. For example, embedding map vocab_size to embed_size, than projection map embed_size back to vocab_siz

Projection based on embedding's weight. For example, embedding map vocab_size to embed_size, than projection map embed_size back to vocab_siz

Projection based on embedding's weight. For example, embedding map vocab_size to embed_size, than projection map embed_size back to vocab_siz

Read a piece of data from dataset. Workers in different ranks will read different data.

Returns a state dict representing the read states of the dataset. Example: >>> state = dataset.state_dict() >>> datas