What happens internally when you call model.fit() in Keras?

model.fit builds a training loop that handles batching, forward pass, loss computation, backpropagation, and optimizer updates.

Internally, Keras compiles a training step using tf.function. It iterates over tf.data pipelines, computes forward pass, calculates loss, computes gradients using GradientTape, and applies optimizer updates. Callbacks are triggered at specific hooks.