python - NaN loss output when training DNN with Keras - Stack Overflow

I am trying to do an exercise from Hands-On ML by Aurelien Geron. However when I tried using my own solution and copy and pasting the solution from the answer key, I keep getting this outcome that is not supposed to happen. The loss and val_loss at each epoch is nan and the accuracy is not increasing at all. This exercise is to try and build a DNN with 20 hidden layers on the CIFAR10 dataset.

Here is the code:

(X_train, y_train), (X_test, y_test) = tf.keras.datasets.cifar10.load_data()
X_val, y_val = X_train[:5000], y_train[:5000]
X_train, y_train = X_train[5000:], y_train[5000:]
tf.random.set_seed(42)

 model = tf.keras.Sequential()
model.add(tf.keras.layers.Input(shape=[32,32,3]))
model.add(tf.keras.layers.Flatten())

## 20 Hidden layers 
for _ in range(20):
    model.add(tf.keras.layers.Dense(100, activation='swish', 
    kernel_initializer='he_normal'))

# Output layer
model.add(tf.keras.layers.Dense(10, activation='softmax'))

optimizer = tf.keras.optimizers.Nadam(learning_rate=5e-6)
modelpile(optimizer=opimizer,
         loss='sparse_categorical_crossentropy',
         metrics=['accuracy'])
import datetime

early_stopping = tf.keras.callbacks.EarlyStopping(patience=20, 
                                   restore_best_weights=True)

log_dir = "logs/my_cifar10_model/" + datetime.datetime.now().strftime("%Y%m%d- 
                  %H%M%S")
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir)

model.fit(X_train, y_train, epochs=100, callbacks=[early_stopping, 
     tensorboard_callback],
     validation_data=(X_val, y_val))

And here is the output:

Epoch 1/100
1407/1407 ━━━━━━━━━━━━━━━━━━━━ 8s 3ms/step - accuracy: 0.1005 - loss: nan - 
val_accuracy: 0.0996 - val_loss: nan
Epoch 2/100
1407/1407 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.1007 - loss: nan - 
val_accuracy: 0.0996 - val_loss: nan
Epoch 3/100
1407/1407 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.1007 - loss: nan - 
val_accuracy: 0.0996 - val_loss: nan

...

Epoch 20/100
1407/1407 ━━━━━━━━━━━━━━━━━━━━ 3s 2ms/step - accuracy: 0.1007 - loss: nan - 
val_accuracy: 0.0996 - val_loss: nan
Epoch 21/100
1407/1407 ━━━━━━━━━━━━━━━━━━━━ 5s 3ms/step - accuracy: 0.1007 - loss: nan - 
val_accuracy: 0.0996 - val_loss: nan
<keras.src.callbacks.history.History at 0x7801cde19fa0>

There are no actually errors rasied by TF or python.

How to resolve this?