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

Logs the metric dict passed in. :param metrics: :param grad_norm_dic:

r"""Applies each `module` in :attr:`modules` in parallel on arguments contained in :attr:`inputs` (positional) and :attr:`kwargs_tup` (keyword)

Reduces output according to the training mode. Separates loss from logging and tqdm metrics :param output: :return:

remove empty lines

Remove weight normalization module from all of the layers.

Sanity check a few things before starting actual training. :param model:

Strip ids_to_strip from the end ids.

Initialize STFT loss module.

Initialize CausalConv1d module.

Saves a checkpoint atomically, avoiding the creation of incomplete checkpoints. This will create a temporary checkpoint with a suffix of ``.p

r""" Recursively find all tensors contained in the specified object.

Only supports IntervalTier currently

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

:param sample: :param batch_idx: :return: total loss: torch.Tensor, loss_log: dict

:param outputs: :return: loss_output: dict

:param dur_input: [B, T_txt, H] :param mel2ph: [B, T_mel] :param txt_tokens: [B, T_txt] :param ret: :return:

Apply weight normalization module from all of the layers.

Apply weight normalization module from all of the layers.

Apply weight normalization module from all of the layers.

Apply weight normalization module from all of the layers.

Apply weight normalization module from all of the layers.

Compute the value. When current unit is larger than previous unit, current unit * all previous units will be used as all previous uni