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

Method forward

model/modeling_bert.py:1068–1079  ·  view source on GitHub ↗
(self, input_ids, token_type_ids=None, attention_mask=None, masked_lm_labels=None,
                checkpoint_activations=False)

Source from the content-addressed store, hash-verified

1066 self.apply(self.init_bert_weights)
1067
1068 def forward(self, input_ids, token_type_ids=None, attention_mask=None, masked_lm_labels=None,
1069 checkpoint_activations=False):
1070 sequence_output, _ = self.bert(input_ids, token_type_ids, attention_mask,
1071 output_all_encoded_layers=False, checkpoint_activations=checkpoint_activations)
1072 prediction_scores = self.cls(sequence_output)
1073
1074 if masked_lm_labels is not None:
1075 loss_fct = CrossEntropyLoss(ignore_index=-1)
1076 masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), masked_lm_labels.view(-1))
1077 return masked_lm_loss
1078 else:
1079 return prediction_scores
1080
1081
1082class BertForNextSentencePrediction(PreTrainedBertModel):

Callers

nothing calls this directly

Calls

no outgoing calls

Tested by

no test coverage detected