Saving hybrid tf.keras nn with gpflux.layers.GPLayer

[ems1111]

I can not figure out how to save the model and reload it wit the same performance as it had when i trained the model
Can someone comment on how i might save this and then reload it i? I feel like it should be more straightforward than it is.

CODE:

Ensure inducing points have the correct dimensions

num_inducing = 2500
inducing_variable = gpflow.inducing_variables.InducingPoints(np.random.randn(num_inducing, 62)) # Match the dense layer output

Define kernel

kernel = gpflow.kernels.SquaredExponential()*gpflow.kernels.White(variance=0.01) + gpflow.kernels.White(variance=0.01)

Define likelihood

likelihood = gpflow.likelihoods.Gaussian(0.01)
likelihood_container = gpflux.layers.TrackableLayer()
likelihood_container.likelihood = likelihood

Hybrid model with explicit layer names

gp_hybrid_model = tf.keras.Sequential([
tf.keras.layers.Dense(156, activation="relu", kernel_initializer='he_normal', bias_initializer='he_normal', name="dense_gp_hybrid_1"),
tf.keras.layers.Dropout(0.35),
tf.keras.layers.Dense(62, activation="relu", kernel_initializer='he_normal', bias_initializer='he_normal', name="dense_gp_hybrid_2"),
gpflux.layers.GPLayer(
kernel=kernel,
inducing_variable=inducing_variable,
num_data=X_train.shape[0],
num_latent_gps=31,
mean_function=gpflow.mean_functions.Zero(),
name="gp_layer"
),
likelihood_container
])

Build the model to initialize the layers

gp_hybrid_model.build(input_shape=(None, 78))