JAXNew

JAX is Google's open-source, NumPy-style numerical computing library. It traces Python functions and compiles them with XLA to CPU/GPU/TPU, and offers composable functional transformations — grad (automatic differentiation), jit (compilation), pmap/vmap (parallelism and vectorization).

JAX is the numerical substrate behind much of modern physics AI: differentiable solvers such as JAX-CFD and JAX-Fluids, neural operators, and PINN/neural-ODE research all build on its autodiff-plus-JIT foundation.