XLA (Accelerated Linear Algebra) is a compiler infrastructure that takes models from ML frameworks such as PyTorch, TensorFlow, and JAX and generates optimized code for execution on GPUs, CPUs, and specialized ML accelerators. It performs graph-level optimizations and hardware-specific code generation to improve performance. The project is the core of the OpenXLA ecosystem and is primarily intended for compiler contributors and integrators adding support for new ML frontends or hardware backends.