I tried saving and loading a tensorflow model on the same device and got the following warning:

UserWarning: Skipping variable loading for optimizer 'adam', because it has 14 variables whereas the saved optimizer has 2 variables.

The model loaded did not behave the same way as when being trained and had a much higher error rate.

Here is the model class:

class DQNAgent:
    def __init__(self, state_size, action_size):
        self.state_size = state_size
        self.action_size = action_size
        self.memory = deque(maxlen = 20000)

        #Other parameters...

        self.model = self._build_model()

    def _build_model(self):
        model = tf.keras.Sequential([
            tf.keras.layers.Dense(64, activation="relu", input_shape=(self.state_size,)),
            tf.keras.layers.Dense(64, activation="relu"),
            tf.keras.layers.Dense(self.action_size, activation=None)
        ])
        model.compile(optimizer = tf.keras.optimizers.Adam(learning_rate = self.alpha), loss = 'mse')
        return model
    
    ...

    def save(self, filepath):
        self.model.save(filepath)

    def load(self, filepath):
        self.model = tf.keras.models.load_model(filepath)

I initialised, then saved the model as such:

agent = DQNAgent(state_size=5, action_size=2)
agent.save("agent.keras")

And loaded it as such:

agent = DQNAgent(state_size=5, action_size=2)
agent.load("agent.keras")
agent.epsilon = 0