XLA

A Google-originated, open-source compiler ('Accelerated Linear Algebra') that takes a machine-learning program expressed as linear-algebra operations and rewrites it into efficient code for a particular chip. Frameworks lower a model's computational graphcomputational graph, n.: A map of a calculation drawn as boxes and arrows: each box is a simple operation and each arrow carries a number to the next step. AI software uses it to trace errors backward. to XLA's intermediate representation, HLO, or its portable, versioned form StableHLO; XLA then performs whole-program optimizations — operator fusion, layout assignment, buffer reuse, algebraic simplification and automatic partitioning of computations across devices — and generates code for CPUs, GPUs and TPUsTPU, abbr.: Google's own family of chips built specifically for AI, used to train and run its Gemini models and rented to other companies through its cloud.. It is the standard compiler path to TPUs and the backbone of JAXJAX, prop. n.: A free Google library that lets researchers write mathematics in familiar Python and automatically get derivatives and very fast code for graphics chips and Google's AI chips.; TensorFlowTensorFlow, prop. n.: Google's free software toolkit for building, training and running machine-learning models; it led the field after 2015 and still runs many models inside phones, browsers and company systems. and PyTorchPyTorch, prop. n.: The most widely used free software toolkit for building and training neural networks, especially in research; programmers write ordinary Python and it handles the calculus and the graphics chips. (through PyTorch/XLA) can also use it. Because it compiles ahead of execution for fixed tensor shapes, dynamic shapes and data-dependent control flow require recompilation or special handling.