Source code for src.training.subclasses.multilabel_distilbert_for_sequence_classification

from torch import nn

from transformers import DistilBertForSequenceClassification
from transformers.modeling_outputs import SequenceClassifierOutput

from ..losses import AsymmetricLossOptimized




class DistilBertForMultiLabelClassification(DistilBertForSequenceClassification):
    """
    Custom implementation for multilabel distilbert classification
    """
    mlb_losses = {
        "asl": AsymmetricLossOptimized,
        "bce": nn.BCEWithLogitsLoss,
    }

    def __init__(self, config, loss='bce', loss_args=None):
        super().__init__(config)
        if loss_args is None:
            loss_args = {}
        self.loss_fct = self.mlb_losses[loss](**loss_args)



    def forward(
            self,
            input_ids=None,
            attention_mask=None,
            head_mask=None,
            inputs_embeds=None,
            labels=None,
            output_attentions=None,
            output_hidden_states=None,
            return_dict=None,
    ):
        """
        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
            Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[0, ...,
            config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
            If :obj:`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

        distilbert_output = self.distilbert(
            input_ids=input_ids,
            attention_mask=attention_mask,
            head_mask=head_mask,
            inputs_embeds=inputs_embeds,
            output_attentions=output_attentions,
            output_hidden_states=output_hidden_states,
            return_dict=return_dict,
        )
        hidden_state = distilbert_output[0]  # (bs, seq_len, dim)
        pooled_output = hidden_state[:, 0]  # (bs, dim)
        pooled_output = self.pre_classifier(pooled_output)  # (bs, dim)
        pooled_output = nn.ReLU()(pooled_output)  # (bs, dim)
        pooled_output = self.dropout(pooled_output)  # (bs, dim)
        logits = self.classifier(pooled_output)  # (bs, dim)

        loss = None
        if labels is not None:
            if self.num_labels == 1:
                loss_fct = nn.MSELoss()
                loss = loss_fct(logits.view(-1), labels.view(-1))
            else:
                loss = self.loss_fct(logits.view(-1, self.num_labels), labels.view(-1, self.num_labels))

        if not return_dict:
            output = (logits,) + distilbert_output[1:]
            return ((loss,) + output) if loss is not None else output

        return SequenceClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=distilbert_output.hidden_states,
            attentions=distilbert_output.attentions,
        )


    forward.__doc__ = DistilBertForSequenceClassification.forward.__doc__