modeling_alignscore.py

from dataclasses import dataclass
from typing import Optional, Tuple, Union

import torch
from torch import nn
from transformers.modeling_outputs import SequenceClassifierOutput
from transformers.models.roberta import RobertaModel, RobertaPreTrainedModel
from .configuration_alignscore import AlignscoreConfig


@dataclass
class ModelOutput:
    loss: Optional[torch.FloatTensor] = None
    all_loss: Optional[list] = None
    loss_nums: Optional[list] = None
    prediction_logits: torch.FloatTensor = None
    seq_relationship_logits: torch.FloatTensor = None
    tri_label_logits: torch.FloatTensor = None
    reg_label_logits: torch.FloatTensor = None
    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
    attentions: Optional[Tuple[torch.FloatTensor]] = None


class AlignscoreModel(RobertaPreTrainedModel):
    config_class = AlignscoreConfig
    # COPIED FROM transformers.models.roberta.modeling_roberta.RobertaForSequenceClassification

    def __init__(self, config):
        super().__init__(config)
        # NUM_LABELS WILL BE IGNOREDD
        # self.num_labels = config.num_labels

        self.config = config

        self.roberta = RobertaModel(config, add_pooling_layer=True)
        self.bin_layer = nn.Linear(config.hidden_size, 2)
        self.tri_layer = nn.Linear(config.hidden_size, 3)
        self.reg_layer = nn.Linear(config.hidden_size, 1)

        if config.hidden_dropout_prob != 0.1:
            print(
                "Warning: The hidden_dropout_prob is not set to 0.1, which may affect the model's performance."
            )
        self.dropout = nn.Dropout(config.hidden_dropout_prob)  # should be 0.1
        self.softmax = nn.Softmax(dim=-1)
        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.FloatTensor] = None,
        token_type_ids: Optional[torch.LongTensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        head_mask: Optional[torch.FloatTensor] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        labels: Optional[torch.LongTensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple[torch.Tensor], 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.roberta(
            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,
        )

        seq_relationship_score = self.bin_layer(
            self.dropout(outputs.pooler_output)
        )  ## pooled output for classification
        tri_label_score = self.tri_layer(self.dropout(outputs.pooler_output))
        reg_label_score = self.reg_layer(outputs.pooler_output)

        if labels is not None:
            raise NotImplementedError(
                "AlignscoreModel does not support labels for training. "
                "Please use the model for inference only."
            )

        return ModelOutput(
            loss=None,
            all_loss=None,
            loss_nums=None,
            prediction_logits=None,
            seq_relationship_logits=seq_relationship_score,
            tri_label_logits=tri_label_score,
            reg_label_logits=reg_label_score,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )