MCPcopy Create free account
hub / github.com/THUDM/GLM / right_strip_seq

Method right_strip_seq

data_utils/datasets.py:672–682  ·  view source on GitHub ↗
(self, tokens, loss_mask, seq_length)

Source from the content-addressed store, hash-verified

670 return {'text': np.array(tokens), "loss_mask": np.array(loss_mask)}
671
672 def right_strip_seq(self, tokens, loss_mask, seq_length):
673 strip_right_tokens = len(tokens) - seq_length
674 if strip_right_tokens > 0:
675 while strip_right_tokens < len(tokens) - 1 and not self.contains_sentence_end(
676 tokens[-strip_right_tokens - 1]):
677 strip_right_tokens += 1
678 if len(tokens) - strip_right_tokens < seq_length // 2:
679 strip_right_tokens = len(tokens) - seq_length
680 tokens = tokens[:-strip_right_tokens]
681 loss_mask = loss_mask[:-strip_right_tokens]
682 return tokens, loss_mask
683
684 def getidx(self, data_idx):
685 data = self.ds[data_idx]

Callers 1

__getitem__Method · 0.95

Calls 1

contains_sentence_endMethod · 0.95

Tested by

no test coverage detected