model.py

from transformers import PreTrainedModel, PretrainedConfig
import torch.nn as nn
from torchvision.models import efficientnet_v2_s, EfficientNet_V2_S_Weights

class CheckboxConfig(PretrainedConfig):
    model_type = "checkbox-classifier"
    
    def __init__(self, num_labels=2, dropout_rate=0.3, **kwargs):
        super().__init__(num_labels=num_labels, **kwargs)
        self.dropout_rate = dropout_rate

class CheckboxClassifier(PreTrainedModel):
    config_class = CheckboxConfig
    
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        
        self.backbone = efficientnet_v2_s(weights=EfficientNet_V2_S_Weights.IMAGENET1K_V1)
        num_features = self.backbone.classifier[1].in_features
        
        self.backbone.classifier = nn.Sequential(
            nn.Dropout(config.dropout_rate),
            nn.Linear(num_features, 512),
            nn.SiLU(inplace=True),
            nn.BatchNorm1d(512),
            nn.Dropout(config.dropout_rate),
            nn.Linear(512, 256),
            nn.SiLU(inplace=True),
            nn.BatchNorm1d(256),
            nn.Dropout(config.dropout_rate/2),
            nn.Linear(256, config.num_labels)
        )
    
    def forward(self, pixel_values):
        outputs = self.backbone(pixel_values)
        return {"logits": outputs}