|
|
from dataclasses import dataclass |
|
|
from typing import Optional, Tuple, Union |
|
|
|
|
|
import torch |
|
|
from torch import nn |
|
|
from torch.nn import BCEWithLogitsLoss, MSELoss |
|
|
from transformers.modeling_outputs import ModelOutput |
|
|
from transformers.models.deberta_v2.modeling_deberta_v2 import ( |
|
|
ContextPooler, DebertaV2Model, DebertaV2PreTrainedModel, StableDropout) |
|
|
|
|
|
|
|
|
@dataclass |
|
|
class SequenceClassifierOutput(ModelOutput): |
|
|
loss: Optional[torch.FloatTensor] = None |
|
|
logits: torch.FloatTensor = None |
|
|
binary_logits: torch.FloatTensor = None |
|
|
hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None |
|
|
attentions: Optional[Tuple[torch.FloatTensor, ...]] = None |
|
|
|
|
|
class DebertaV2ForSequenceClassification(DebertaV2PreTrainedModel): |
|
|
def __init__(self, config): |
|
|
super().__init__(config) |
|
|
self.deberta = DebertaV2Model(config) |
|
|
self.pooler = ContextPooler(config) |
|
|
output_dim = self.pooler.output_dim |
|
|
|
|
|
self.binary_classifier = nn.Linear(output_dim, 1) |
|
|
self.regressor = nn.Linear(output_dim, 1) |
|
|
drop_out = getattr(config, "cls_dropout", None) |
|
|
drop_out = self.config.hidden_dropout_prob if drop_out is None else drop_out |
|
|
self.dropout = StableDropout(drop_out) |
|
|
self.post_init() |
|
|
|
|
|
def freeze_embeddings(self) -> None: |
|
|
"""Frezees the embedding layer.""" |
|
|
for param in self.deberta.embeddings.parameters(): |
|
|
param.requires_grad = False |
|
|
|
|
|
def get_input_embeddings(self): |
|
|
return self.deberta.get_input_embeddings() |
|
|
|
|
|
def set_input_embeddings(self, new_embeddings): |
|
|
self.deberta.set_input_embeddings(new_embeddings) |
|
|
|
|
|
def forward( |
|
|
self, |
|
|
input_ids: Optional[torch.Tensor] = None, |
|
|
attention_mask: Optional[torch.Tensor] = None, |
|
|
token_type_ids: Optional[torch.Tensor] = None, |
|
|
position_ids: Optional[torch.Tensor] = None, |
|
|
inputs_embeds: Optional[torch.Tensor] = None, |
|
|
labels: Optional[torch.Tensor] = None, |
|
|
output_attentions: Optional[bool] = None, |
|
|
output_hidden_states: Optional[bool] = None, |
|
|
return_dict: Optional[bool] = None, |
|
|
) -> Union[Tuple, SequenceClassifierOutput]: |
|
|
r""" |
|
|
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*): |
|
|
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ..., |
|
|
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If |
|
|
`config.num_labels > 1` a classification loss is computed (Cross-Entropy). |
|
|
""" |
|
|
return_dict = return_dict if return_dict is not None else self.config.use_return_dict |
|
|
|
|
|
outputs = self.deberta( |
|
|
input_ids, |
|
|
token_type_ids=token_type_ids, |
|
|
attention_mask=attention_mask, |
|
|
position_ids=position_ids, |
|
|
inputs_embeds=inputs_embeds, |
|
|
output_attentions=output_attentions, |
|
|
output_hidden_states=output_hidden_states, |
|
|
return_dict=return_dict, |
|
|
) |
|
|
|
|
|
encoder_layer = outputs[0] |
|
|
pooled_output = self.pooler(encoder_layer) |
|
|
pooled_output = self.dropout(pooled_output) |
|
|
binary_logits = self.binary_classifier(pooled_output) |
|
|
logits = self.regressor(pooled_output) |
|
|
|
|
|
loss = None |
|
|
if labels is not None: |
|
|
regression_loss_fct = MSELoss() |
|
|
regression_loss = regression_loss_fct(logits.squeeze(), labels.squeeze().float()) |
|
|
|
|
|
binary_loss_fct = BCEWithLogitsLoss() |
|
|
|
|
|
binary_labels = (labels >= 3).float() |
|
|
classification_loss = binary_loss_fct(binary_logits.squeeze(), binary_labels.squeeze()) |
|
|
loss = regression_loss + classification_loss |
|
|
|
|
|
if not return_dict: |
|
|
output = (logits,) + outputs[1:] |
|
|
return ((loss,) + output) if loss is not None else output |
|
|
|
|
|
return SequenceClassifierOutput( |
|
|
loss=loss, logits=logits, binary_logits=binary_logits, hidden_states=outputs.hidden_states, attentions=outputs.attentions |
|
|
) |
