model.py

import torch
from torch import nn
from torchvision.models import EfficientNet_B2_Weights, efficientnet_b2

def create_effnetb2_model(num_classes: int = 3,
                          seed: int = 42):
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)

    weights = EfficientNet_B2_Weights.DEFAULT
    transforms = weights.transforms()
    model = efficientnet_b2(weights=weights)

    for param in model.parameters():
        param.requires_grad = False
    
    model.classifier[1] = nn.Linear(in_features=1408, out_features=num_classes)

    return model, transforms