Added BetForEnformer class

modeling_bert.py

@@ -2206,3 +2206,87 @@ def rotate_half(x):
 def apply_rotary_pos_emb(q, k, cos, sin, offset: int = 0):
     cos, sin = cos[:, :, offset: q.shape[2] + offset, :], sin[:, :, offset: q.shape[2] + offset, :]
     return (q * cos) + (rotate_half(q) * sin), (k * cos) + (rotate_half(k) * sin)
+    
+    
+from torch import nn
+from transformers.modeling_outputs import TokenClassifierOutput
+
+class BertForEnformer(BertPreTrainedModel):
+
+    _keys_to_ignore_on_load_unexpected = [r"pooler"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.bert = BertModel(config, add_pooling_layer=False)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
+        self.activation = nn.Softplus()
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        bins_mask=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        labels_mask=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+
+        # select SEP tokens that represent target bins
+        bins_output = sequence_output[bins_mask]
+
+        bins_output = self.dropout(bins_output)
+        logits = self.classifier(bins_output)
+        pred = self.activation(logits)
+
+        loss = None
+        if labels is not None:
+            loss_fct = nn.PoissonNLLLoss(log_input=False, reduction='mean')
+            labels = labels[labels_mask]
+            loss = loss_fct(pred, labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+