Зависает после повторного расчета

stacked_encoder = Sequential([
    Flatten(input_shape=[lenght_axis1,lenght_axis2,lenght_axis3]),
    Dense(100, activation="selu"),
    Dense(10, activation="selu"),
])
 
stacked_encoder.summary()
 
stacked_decoder = Sequential([
    Dense(100, activation="selu", input_shape=[10]),
    Dense(331776, activation="sigmoid"),
    Reshape([lenght_axis1,lenght_axis2,lenght_axis3])
])
stacked_decoder.summary()
 
stacked_ae = Sequential([stacked_encoder, stacked_decoder])
stacked_ae.summary()

import matplotlib.pyplot as plt
import numpy as np
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Flatten, Dense, Reshape
from tensorflow.keras.optimizers import SGD
from tensorflow.keras.metrics import binary_accuracy
from tensorflow.math import round
 
if __name__ == '__main__':
    img_origin = plt.imread(r'...') #Размер изображения (311, 653, 3)
 
    new_height, new_width = 288, 384
    new_stride = 4
 
    dataSet = np.lib.stride_tricks.sliding_window_view(img_origin, window_shape=(new_height, new_width, 3))[
              ::new_stride, ::new_stride, :, :, :, :].reshape(-1, new_height, new_width, 3)
    
    dataSet_norm =  (dataSet/255.0 * 0.99) + 0.01  - 0.5
 
    border = int(dataSet_norm.shape[0] * 0.8)
    x_train = dataSet_norm[:border]
    x_test  = dataSet_norm[border:]
 
    def rounded_accuracy(y_true, y_pred):
        return binary_accuracy(round(y_true), round(y_pred))
 
    #Зависает тут
    #----------------------------------------------------------------------------------
    stacked_encoder = Sequential([
        Flatten(input_shape=[dataSet.shape[1], dataSet.shape[2], 3]),
        Dense(1000, activation="selu"),
        Dense(100, activation="selu"),
        Dense(10, activation="selu"),
    ])
 
    stacked_encoder.summary()
 
    stacked_decoder = Sequential([
        Dense(100, activation="selu", input_shape=[10]),
        Dense(1000, activation="selu"),
        Dense(dataSet.shape[1] * dataSet.shape[2] * 3, activation="sigmoid"),
        Reshape([dataSet.shape[1], dataSet.shape[2], 3])
    ])
    stacked_decoder.summary()
 
    stacked_ae = Sequential([stacked_encoder, stacked_decoder])
    stacked_ae.summary()
    #----------------------------------------------------------------------------------