KerasTuner in ImageClassification problem

I am doing image classification. I was trying classical methods (GridSearch, Pipeline) for finding the best parameters, but as I imported my images with ImageDataGenerator it was not working very well.

In the searches for a solution, I found KerasTuner (never heard of this before). The code examples I found on a website seemed to be not hard to implement. So I tried it. But it is taking too long for just a training evaluation (not any result of test data). So I thought maybe I did something wrong in my script (see below)

It is running already for the second day on 2080Ti, and I don't have even val_accuracy. For accuracy I am already having 0.9736 and loss:0.0604 which is great, to be honest. But will it take a lifetime until a real result?

My dataset consists of 12000 images for training and 1600 for testing. (4 classes)

from kerastuner.applications import HyperResNet
from kerastuner.tuners import Hyperband
from keras.preprocessing.image import ImageDataGenerator

train_datagen = ImageDataGenerator(
        rescale=1./255,
        horizontal_flip=True
)

# augmentation configuration for testing, only rescale
test_datagen = ImageDataGenerator(rescale=1./255)

batch_size = 128

train_generator = train_datagen.flow_from_directory(
        '/media/data/working_dir/categories/readytotest/train',
        target_size=(240, 320),
        color_mode="grayscale",
        batch_size=batch_size,
        shuffle=True,
        class_mode='categorical')

validation_generator = test_datagen.flow_from_directory(
        '/media/data/working_dir/categories/readytotest/test',
        target_size=(240, 320),
        color_mode="grayscale",
        batch_size=batch_size,
        shuffle=True,
        class_mode='categorical')

hypermodel = HyperResNet(input_shape=(240, 320, 1), classes=4)

tuner = Hyperband(
    hypermodel,
    objective='val_accuracy',
    max_epochs=5,
    directory='/media/data/working_dir/scr/models/',
    project_name='first_kt')

print("SUMMARY OF THE TASK")
tuner.search_space_summary()

# perform analyses
N_EPOCH_SEARCH = 5
tuner.search(train_generator,
             epochs=N_EPOCH_SEARCH,
             validation_data=validation_generator)
``