linear_crf

class hanlp.components.mtl.tasks.pos.LinearCRFDecoder(hidden_size, num_labels, crf=False)[source]

A linear layer with an optional CRF (Lafferty et al. 2001) layer on top of it.

Parameters
  • hidden_size – Size of hidden states.

  • num_labels – Size of tag set.

  • crfTrue to enable CRF (Lafferty et al. 2001).

forward(contextualized_embeddings: torch.FloatTensor, batch: Dict[str, torch.Tensor], mask=None)[source]
Parameters
  • contextualized_embeddings – Hidden states for contextual layer.

  • batch – A dict of a batch.

  • mask – Mask for tokens.

Returns

Logits. Users are expected to call CRF.decode on these emissions during decoding and CRF.forward during training.