Transform a human-readable string into a sequence of int ids. The ids should be in the range [num_reserved_ids, vocab_size). Ids [0,

Function to reshape the time vector t by the number of dimensions of x. Parameters ---------- x : Tensor, shape (bs, *dim) repres

Helper for setting incremental state for an nn.Module.

Helper for getting incremental state for an nn.Module.

Initialize vocabulary with tokens from token_generator.

将中文文本分成多个块，优先确保每个块以句号、感叹号或问号结尾， 其次考虑逗号等其他标点符号，避免在无标点处断句 参数: text: 要分块的文本 limit: 每个块的中文字符数限制 look_ahead_limit

Splits the input text into chunks, each with a maximum number of characters. Args: text (str): The text to be split. max_cha

Compute the standard deviation of the probability path N(t * x1 + (1 - t) * x0, sigma), see (Eq.14) [1]. Parameters --------

Build sinusoidal embeddings. This matches the implementation in tensor2tensor, but differs slightly from the description in Section 3

Build sinusoidal embeddings. This matches the implementation in tensor2tensor, but differs slightly from the description in Section 3

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

x: [b, h, l, l] ret: [b, h, l, 2*l-1]

Bias for self-attention to encourage attention to close positions. Args: length: an integer scalar. Returns: a Ten

When we use torchdiffeq, we need to include the CFG process inside _forward()

Load vocab from a file. Args: filename: The file to load vocabulary from.

Initialize tokens from a list of tokens. It is ok if reserved tokens appear in the vocab list. They will be removed. The set of token

x: [b, h, l, d] y: [h or 1, m, d] ret: [b, h, l, m]

x: [b, h, l, m] y: [h or 1, m, d] ret: [b, h, l, d]

x: [b, h, l, 2*l-1] ret: [b, h, l, l]

Rotate the input tensor by half of its innermost width. input (Tensor): array to rotate. RETURNS (Tensor): rotated array. Sh

Compute the conditional vector field ut(x1|x0) = x1 - x0, see Eq.(15) [1]. Parameters ---------- x0 : Tensor, shape

Compute the mean of the probability path N(t * x1 + (1 - t) * x0, sigma), see (Eq.14) [1]. Parameters ---------- x0

Transform a sequence of int ids into a human-readable string. EOS is not expected in ids. Args: ids: list of integers to be

Transform a sequence of int ids into a their string versions. This method supports transforming individual input/output ids to their

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

:param x: [B, T, C] :param padding_mask: [B, T] :return: [B, T, C] or [L, B, T, C]