Deep learning frameworks and compilers: from NumPy to PyTorch and JAX

Discuss where machine learning frameworks are heading: how they moved from NumPy to PyTorch and JAX, and how PyTorch and JAX differ (eager versus traced execution, functional transforms, compilation). Then trace what happens when a model runs on a GPU: graph capture, compiler optimizations such as operator fusion and memory planning, kernel selection and launch, and how software is matched to the hardware.