Update bert_layers.py
Browse files- bert_layers.py +4 -3
bert_layers.py
CHANGED
@@ -491,6 +491,7 @@ class BertEncoder(nn.Module):
|
|
491 |
if not output_all_encoded_layers:
|
492 |
all_encoder_layers.append(hidden_states)
|
493 |
return all_encoder_layers, all_attention_weights # Return both hidden states and attention weights
|
|
|
494 |
|
495 |
|
496 |
|
@@ -614,13 +615,13 @@ class BertModel(BertPreTrainedModel):
|
|
614 |
first_col_mask[:, 0] = True
|
615 |
subset_mask = masked_tokens_mask | first_col_mask
|
616 |
|
617 |
-
encoder_outputs
|
618 |
embedding_output,
|
619 |
attention_mask,
|
620 |
output_all_encoded_layers=output_all_encoded_layers,
|
621 |
subset_mask=subset_mask)
|
622 |
-
print(f'here is the matrix of attentions in BERT: \n {all_attention_weights}')
|
623 |
-
print(f'and this is the shape in BERT: \n {all_attention_weights.shape}')
|
624 |
|
625 |
if masked_tokens_mask is None:
|
626 |
sequence_output = encoder_outputs[-1]
|
|
|
491 |
if not output_all_encoded_layers:
|
492 |
all_encoder_layers.append(hidden_states)
|
493 |
return all_encoder_layers, all_attention_weights # Return both hidden states and attention weights
|
494 |
+
return all_encoder_layers # Return both hidden states and attention weights
|
495 |
|
496 |
|
497 |
|
|
|
615 |
first_col_mask[:, 0] = True
|
616 |
subset_mask = masked_tokens_mask | first_col_mask
|
617 |
|
618 |
+
encoder_outputs = self.encoder(
|
619 |
embedding_output,
|
620 |
attention_mask,
|
621 |
output_all_encoded_layers=output_all_encoded_layers,
|
622 |
subset_mask=subset_mask)
|
623 |
+
# print(f'here is the matrix of attentions in BERT: \n {all_attention_weights}')
|
624 |
+
# print(f'and this is the shape in BERT: \n {all_attention_weights.shape}')
|
625 |
|
626 |
if masked_tokens_mask is None:
|
627 |
sequence_output = encoder_outputs[-1]
|