Update modeling_bert.py

modeling_bert.py

@@ -27,12 +27,12 @@ from packaging import version
 from torch import nn
 from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
 
-from ...activations import ACT2FN
-from ...modeling_attn_mask_utils import (
+from transformers.activations import ACT2FN
+from transformers.modeling_attn_mask_utils import (
     _prepare_4d_attention_mask_for_sdpa,
     _prepare_4d_causal_attention_mask_for_sdpa,
 )
-from ...modeling_outputs import (
+from transformers.modeling_outputs import (
     BaseModelOutputWithPastAndCrossAttentions,
     BaseModelOutputWithPoolingAndCrossAttentions,
     CausalLMOutputWithCrossAttentions,
@@ -43,9 +43,9 @@ from ...modeling_outputs import (
     SequenceClassifierOutput,
     TokenClassifierOutput,
 )
-from ...modeling_utils import PreTrainedModel
-from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
-from ...utils import (
+from transformers.modeling_utils import PreTrainedModel
+from transformers.pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
+from transformers.utils import (
     ModelOutput,
     add_code_sample_docstrings,
     add_start_docstrings,
@@ -56,8 +56,7 @@ from ...utils import (
 )
 from .configuration_bert import BertConfig
 
-BertModel.register_for_auto_class("AutoModel")
-BertForMaskedLM.register_for_auto_class("AutoModelForMaskedLM")
+
 def softmax_n_shifted_zeros(input: torch.Tensor, n: int, dim=-1) -> torch.Tensor:
     """
     $\text(softmax)_n(x_i) = exp(x_i) / (n + \sum_j exp(x_j))$
@@ -2282,3 +2281,5 @@ class BertForQuestionAnswering(BertPreTrainedModel):
             hidden_states=outputs.hidden_states,
             attentions=outputs.attentions,
         )
+BertModel.register_for_auto_class("AutoModel")
+BertForMaskedLM.register_for_auto_class("AutoModelForMaskedLM")