How to get loss gradient wrt internal layer output in tensorflow 2?

In TensorFlow 2, you can use the tf.GradientTape context manager to compute the gradient of a loss with respect to the output of an internal layer. Here's an example of how you can do this:

import tensorflow as tf

# Build the model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, input_shape=(input_shape,), activation='relu'),
    tf.keras.layers.Dense(1)
])

# Compile the model with a loss function and an optimizer
model.compile(loss='mean_squared_error', optimizer='adam')

# Generate some fake data for training
x_train = np.random.random((100, input_shape))
y_train = np.random.random((100, 1))

# Use the model to predict on the training data
with tf.GradientTape() as tape:
    logits = model(x_train, training=True)
    loss_value = tf.reduce_mean(tf.square(logits - y_train))

# Use the tape to compute the gradient of the loss with respect to the output of the internal layer
gradients = tape.gradient(loss_value, logits)

# Now you can use the gradients to update the model weights
optimizer.apply_gradients(zip(gradients, model.trainable_variables))

This code will compute the gradient of the loss with respect to the output of the internal layer (i.e., the output of the first dense layer) and store it in the gradients variable. You can then use this gradient to update the model weights using an optimizer, as shown in the last line of the code snippet.