Load a pretrained component from an identifier. Args: save_dir (str): The identifier to the saved component. It could be a remote URL or a

Load file ids from a Chinese treebank grouped by domains. Args: ctb_home: Root path to CTB. Returns: A dict of sets, ea

Load tokenized corpus. The format is one sentence per line, where each line consisits of tokens seperated by a delimiter (usually space).

Load multi-criteria corpora specified in filepath. Args: filepath: A list of files where filename is its criterion. Or a dict of

Load ``.jsonlines`` NER corpus. Samples of this corpus can be found using the following scripts. .. highlight:: python .. code-block:

Load a ``.tsv`` file. A ``.tsv`` file for tagging is defined as a tab separated text file, where non-empty lines have two columns for token an

Both ``.conllx`` and ``.conllu`` are supported. Their descriptions can be found in :class:`hanlp_common.conll.CoNLLWord` and :class:`hanlp_com

Load ``.jsonlines`` CoNLL12-style corpus. Samples of this corpus can be found using the following scripts. .. highlight:: python .. c

Load eos corpus. Args: filepath: Path to the corpus. .. highlight:: bash .. code-block:: bash $ hea

Load vocabularies from a directory. Args: save_dir: The directory to load vocabularies. filename: The name for vocab

Lock each vocab.

``True`` indicates this vocab is locked.

r""" Args: s_span (~torch.Tensor): ``[batch_size, seq_len, seq_len]``. Scores of all spans s_label (~t

Merge this with another vocab inplace. Args: other (Vocab): Another vocab.

r''' * Takes in two lists of matrices `state_S` and `state_T`. Each list contains 2 matrices of the shape (*batch_size*, *length*, *hidden_size*).

The actual model when it's wrapped by a `DataParallel` Returns: The "real" model

r""" MST algorithm for decoding non-pojective trees. This is a wrapper for ChuLiu/Edmonds algorithm. The algorithm first runs ChuLiu/Edmo