The config of this object, which are public properties. If any properties needs to be excluded from this config, simply declare it wi

Remember to detach logits_S

* Computes the cosine similarity loss between the inputs. This is the loss used in DistilBERT, see `DistilBERT <https://arxiv.org/abs/1910.01108>

Creates an optimizer with learning rate schedule. Args: init_lr: num_train_steps: num_warmup_steps: Returns:

Computes the unnormalized score of all tag sequences matching tag_bitmap. tag_bitmap enables more than one tag to be considered correct a

Convert ctb pos tagging corpus from tsv format to text format, where each word is followed by its pos tag. Args: path: File to be

An iterator over the entire dataset, yielding all sentences processed.

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

Args: span_starts: [num_candidates,] span_scores: [num_candidates, num_labels] span_ends: labels_inv: Returns:

Special json encoder for numpy types See https://interviewbubble.com/typeerror-object-of-type-float32-is-not-json-serializable/ Args:

The devices this component lives on.