import torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PreTrainedModel, PretrainedConfig


class CaptchaConfig(PretrainedConfig):
    model_type = "captcha"  # Importante para o Hugging Face reconhecer

    def __init__(self, input_dim=(40, 110), output_ndigits=5, output_vocab_size=10, vocab=None, **kwargs):
        super().__init__(**kwargs)
        self.input_dim = input_dim
        self.output_ndigits = output_ndigits
        self.output_vocab_size = output_vocab_size
        self.vocab = vocab if vocab else [str(i) for i in range(10)]


class CaptchaModel(PreTrainedModel):
    config_class = CaptchaConfig
    model_type = "captcha"  # Importante para o Hugging Face reconhecer

    def __init__(self, config):
        super().__init__(config)
        self.vocab = config.vocab
        self.output_ndigits = config.output_ndigits
        self.output_vocab_size = config.output_vocab_size

        self.batchnorm0 = nn.BatchNorm2d(3)
        self.conv1 = nn.Conv2d(3, 32, kernel_size=3)
        self.batchnorm1 = nn.BatchNorm2d(32)
        self.conv2 = nn.Conv2d(32, 64, kernel_size=3)
        self.batchnorm2 = nn.BatchNorm2d(64)
        self.conv3 = nn.Conv2d(64, 64, kernel_size=3)
        self.batchnorm3 = nn.BatchNorm2d(64)
        self.dropout1 = nn.Dropout(0.25)
        self.dropout2 = nn.Dropout(0.5)

        def calc_dim(x):
            for _ in range(3):
                x = (x - 2) // 2
            return x

        conv_h = calc_dim(config.input_dim[0])
        conv_w = calc_dim(config.input_dim[1])
        fc1_in_features = conv_h * conv_w * 64

        self.fc1 = nn.Linear(fc1_in_features, 200)
        self.batchnorm_dense = nn.BatchNorm1d(200)
        self.fc2 = nn.Linear(200, self.output_vocab_size * self.output_ndigits)

    def forward(self, pixel_values):
        x = self.batchnorm0(pixel_values)
        x = F.relu(self.batchnorm1(F.max_pool2d(self.conv1(x), 2)))
        x = F.relu(self.batchnorm2(F.max_pool2d(self.conv2(x), 2)))
        x = F.relu(self.batchnorm3(F.max_pool2d(self.conv3(x), 2)))
        x = torch.flatten(x, start_dim=1)
        x = self.dropout1(x)
        x = F.relu(self.batchnorm_dense(self.fc1(x)))
        x = self.dropout2(x)
        logits = self.fc2(x)
        logits = logits.view(-1, self.output_ndigits, self.output_vocab_size)
        return logits