model.py

import torch.nn as nn
from transformers import AutoModel


class BiasClassifier(nn.Module):
    """KoSBI 전용 (15-label)"""
    def __init__(self, model_name, num_labels):
        super().__init__()
        self.encoder    = AutoModel.from_pretrained(model_name, attn_implementation="eager")
        self.dropout    = nn.Dropout(0.2)
        self.classifier = nn.Linear(self.encoder.config.hidden_size, num_labels)

    def forward(self, input_ids, attention_mask, output_attentions=False, **kwargs):
        out = self.encoder(
            input_ids=input_ids,
            attention_mask=attention_mask,
            output_attentions=output_attentions
        )
        logits = self.classifier(self.dropout(out.last_hidden_state[:, 0, :]))
        attentions = out.attentions if output_attentions else None
        return logits, attentions


class KMHaSClassifier(nn.Module):
    """K-MHaS 전용 (8-label)"""
    def __init__(self, model_name, num_labels=8, hidden_size=1024):
        super().__init__()
        self.encoder    = AutoModel.from_pretrained(model_name, attn_implementation="eager")
        self.classifier = nn.Sequential(
            nn.Linear(hidden_size, 512),
            nn.GELU(),
            nn.Dropout(0.1),
            nn.Linear(512, num_labels),
        )

    def forward(self, input_ids, attention_mask, **kwargs):
        out = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
        return self.classifier(out.last_hidden_state[:, 0, :])


class FramingBiasClassifier(nn.Module):
    """BABE 전용 (binary)"""
    def __init__(self, model_name, hidden_size=1024):
        super().__init__()
        self.encoder    = AutoModel.from_pretrained(model_name, attn_implementation="eager")
        self.classifier = nn.Sequential(
            nn.Linear(hidden_size, 256),
            nn.GELU(),
            nn.Dropout(0.1),
            nn.Linear(256, 1),
        )

    def forward(self, input_ids, attention_mask, **kwargs):
        out = self.encoder(input_ids=input_ids, attention_mask=attention_mask)
        return self.classifier(out.last_hidden_state[:, 0, :])