`tape` is required when a `Tensor` loss is passed

Question

Some question about tf.

import numpy as np
import tensorflow as tf
from tensorflow import keras

x_train = [1,2,3]
y_train = [1,2,3]

W = tf.Variable(tf.random.no