Initialize from a file or list, one token per line. Handling of reserved tokens works as follows: - When initializing from a list,

Yield mini-batches of indices bucketed by size. Batches may contain sequences of different lengths. Args: indices (List[int

Convert a list of 1d tensors into a padded 2d tensor.

Convert a list of 2d tensors into a padded 3d tensor.

Decorator to make any fx with this use the lazy property :param fn: :return:

Entry point into a DP thread :param gpu_idx: :param model: :param cluster_obj: :return:

Converts a space-separated string of tokens to a list of ids.

FP16-compatible function that fills a tensor with -inf.

FP16-compatible function that fills a tensor with -inf.

:param x: B * T * 1 :return: x

Input is expected to be of size [bsz x seqlen].

Input shape: Time x Batch x Channel Args: key_padding_mask (ByteTensor, optional): mask to exclude keys that

Args: incremental_state: Used to buffer signal; if not None, then input is expected to contain a single frame.

Apply layer normalization. :param torch.Tensor x: input tensor :return: layer normalized tensor :rtype torch.Tensor

Calculate forward propagation. Args: xs (Tensor): Batch of input sequences (B, Tmax, idim). x_masks (ByteTensor, op

Calculate forward propagation. Args: outputs (Tensor): Batch of prediction durations in log domain (B, T) targets (

Calculate forward propagation. Args: xs (Tensor): Batch of sequences of char or phoneme embeddings (B, Tmax, D). ds

:param x: [B, T, C] :param require_w: True if this module needs to return weight matrix :return: [B, T, C]

:param src_tokens: [B, T] :return: { 'encoder_out': [T x B x C] 'encoder_padding_mask': [B x T]

:param src_tokens: [B, T] :param vid2ph: :param guide_face: [B, img_h, img_w, C] :return: { 'lip_o

attn: bx x L_t x L_s attn_ks: shape: tensor with shape [batch_size], input_lens/output_lens diagonal: y=k*x (k=attn_ks, x:outpu

attn: bs x L_t x L_s

Helper for getting incremental state for an nn.Module.

attn: bs x L_t x L_s

attn: bs x L_t x L_s

Replace non-padding symbols with their position numbers. Position numbers begin at padding_idx+1. Padding symbols are ignored.

Maximum number of supported positions.