Features

Cross-cutting capabilities that span multiple subsystems. Each feature touches the runtime, Python frontend, and often the compiler stack.

Page	What it covers
tf.function and AutoGraph	Tracing Python functions into graphs, AutoGraph source rewriter
tf.data input pipelines	Dataset / Iterator framework for high-throughput input
SavedModel and Checkpoint	The default model serialization formats
Distribution strategy	Multi-GPU / multi-host / TPU training
Quantization	Post-training and during-training quantization for inference
Gradient computation	tf.GradientTape, tf.gradients, registered gradient ops

These pages cross-reference both systems (the runtime they ride on) and compilers (the lowering paths they use).