If we can directly make Sequential in TensorFlow with few lines of code very easily then what is the use of explicit forward propagation and gradient tape?
1 Answers
Well, this is because gradient tape allows you to have much more control on your training. You have a direct access to individual gradients.
With the tape you record the sequence of operations that are performed. This can be useful for creating custom trainings or while working with custom layers for example. You can also decide how to compute the loss directly in your tape.
Sometimes you have simple models and you can directly use compile and fit. Gradient tape is for customization.
Related