TensorFlow results are not reproducible while the global seed is set - Python - Tensorflow - MacOS M1

My results are not reproducible while the global seed is set. Am I doing something wrong or is this an issue of the macbook M1?

I am using:

• MacOS 12.5
• Python 3.10.6
• tensorflow-macos 2.9.2

The terminal is opened using the use rosetta option.

Besides, the model is also unusually bad.

The datasets are from https://www.kaggle.com/competitions/dogs-vs-cats

# Imports.
import tensorflow as tf             
import numpy as np

# Some global variables.
seed = 1234             
batch_size = 32         
learning_rate = 0.001   
width = 180             
height = 180            
epochs = 250            

# Set the seed.
tf.random.set_seed(seed);
np.random.seed(seed);

# Create the cnn model
def cnn_model():
    model = tf.keras.models.Sequential([
        tf.keras.layers.Conv2D(32, (3,3), activation="relu", input_shape=(width, height, 3)),
        tf.keras.layers.MaxPooling2D((2,2)),
        tf.keras.layers.Conv2D(64, (3,3), activation="relu"),
        tf.keras.layers.MaxPooling2D((2,2)),
        tf.keras.layers.Conv2D(128, (3,3), activation="relu"),
        tf.keras.layers.MaxPooling2D((2,2)),
        tf.keras.layers.Conv2D(128, (3,3), activation="relu"),
        tf.keras.layers.MaxPooling2D((2,2)),
        tf.keras.layers.Flatten(),              
        tf.keras.layers.Dropout(0.2),
        tf.keras.layers.Dense(512, activation="relu"),
        tf.keras.layers.Dense(1,    activation="sigmoid"),
    ])
    opt = tf.keras.optimizers.Adam(learning_rate=0.001)
    model.compile(optimizer=opt, loss="binary_crossentropy", metrics=["accuracy"])
    return model

# Load an image for predictions.
def load_image(path):
    img = tf.keras.preprocessing.image.load_img(
        path, 
        target_size=(width, height),
        color_mode="rgb",
    )
    img_tensor = tf.keras.preprocessing.image.img_to_array(img)     
    img_tensor = np.expand_dims(img_tensor, axis=0)                 
    img_tensor /= 255.
    return img_tensor

# Dump the evaluation metrics to the console.
def dump_metrics(message, model, metrics):
    index, s = 0, ""
    for name in model.metrics_names:
        s += f" - {name}: {round(metrics[index], 3)}"
        index += 1
    print(f"{message} {s[3:]}")

# Define the image data generator.
train_datagen = tf.keras.preprocessing.image.ImageDataGenerator(
    rescale = 1./255,
    # rotation_range = 45,
    # width_shift_range = 0.2,      
    # height_shift_range = 0.2,
    # shear_range = 0.2,
    # zoom_range = 0.2
)
val_datagen = tf.keras.preprocessing.image.ImageDataGenerator(rescale = 1./255)

# Load the training dataset.
train_df = train_datagen.flow_from_directory(
    "datasets/train/",              
    class_mode="binary",            
    batch_size=batch_size,          
    target_size=(width, height),    
    shuffle=True,                   
    color_mode="rgb",
    seed=seed,
)
train_samples_count = len(train_df)

# Load the validation dataset.
val_df = val_datagen.flow_from_directory(
    "datasets/val/",                
    class_mode="binary",            
    batch_size=batch_size,          
    target_size=(width, height),    
    shuffle=True,                   
    color_mode="rgb",
    seed=seed,
)
val_samples_count = len(train_df)

# Create the model.
model = cnn_model();
model.summary()                     

# Training callbacks.
callbacks = []
callbacks.append(tf.keras.callbacks.EarlyStopping(monitor="val_loss", patience=5, restore_best_weights=True))

# Train the model.
history = model.fit(
    train_df,                                           
    steps_per_epoch=train_samples_count // batch_size,  
    epochs=epochs,                                      
    validation_data=val_df,                             
    validation_steps=val_samples_count // batch_size,   
    callbacks=callbacks,                                
)
dump_metrics("Training:", model, model.evaluate(train_df, verbose=0))
dump_metrics("Valiation:", model, model.evaluate(val_df, verbose=0))