Thanks for a clear explanation with lots of examples. Can we call a function from within the GradientTape()'s scope?

```python

W1 = tf.Variable(tf.random.normal((f,C)))

b1 = tf.Variable(tf.random.normal((C,)))

def forward_pass(X):

Z1 = X_train @ W1 + b1

A1 = tf.nn.softmax(Z1)

y_hat = A1

return y_hat

for i in range(epochs):

with tf.GradientTape() as tape:

y_hat = forward_pass(X_train)

# versus the inlined impl below

# Z1 = X_train @ W1 + b1

# A1 = tf.nn.softmax(Z1)

# y_hat = A1

J = - tf.reduce_sum( tf.math.log(y_hat) * y_train ) / m

```

I see different results b/w inlining and calling a function