ValueError: No gradients provided for any variable in semi/self supervised loss function

Question

I am training a neural network for clustering applications in a semi/self-supervised way: Instead of having the ground truth, I define the loss function by calculating the similarity among the data points assigned to the same clusters, like:

def loss_function(self, y_true, y_pred):
    def get_loss(x_input, y_input):
        similarity = 0
        for i in range(len(np.unique(y_input))):
            similarity += sum(pdist(x_input))
        return similarity
    score = tf.numpy_function(get_loss, [self.x_input, y_pred], tf.float32)
    return score

In calculating the loss, I don't use y_true, and instead, I use self.x_input, which is the original data point.

I'm getting the following error while running my code:

raise ValueError(f"No gradients provided for any variable: {variable}. "
ValueError: No gradients provided for any variable

So my question is it possible to train a neural network model in this way (without having ground truth)? If so, what is causing the above problem?