JAX
JAX is an open-source numerical computing library from Google that extends NumPy with composable function transformations: automatic differentiation (grad), just-in-time compilation via XLA (jit), vectorization (vmap), and parallelization (pmap). It runs transparently on CPUs, GPUs, and TPUs using a pure-functional programming model. JAX is the low-level engine behind Google's Gemini, Veo, and other frontier AI systems, and is the foundation for higher-level neural network libraries like Flax and Equinox.
JAX has become the dominant framework inside Google DeepMind and is increasingly adopted by top AI research labs for training large-scale models, because its composable transformations enable hardware-efficient gradient computation and distributed training patterns that are difficult to express in PyTorch or TensorFlow. Engineers who understand JAX's functional paradigm, XLA compilation, and the Flax/Optax ecosystem are sought after for frontier model training roles. With Keras 3 now supporting JAX as a backend alongside PyTorch and TensorFlow, JAX proficiency extends the value of deep learning engineers across multiple production stacks.
🎓 Courses
Build and Train an LLM with JAX
by DeepLearning.AI
The most prominent, up-to-date JAX course available — builds a 20M-parameter GPT-2 style language model from scratch, covering jit, grad, vmap, checkpointing, and inference via a chat interface. Free to audit.
Introduction to JAX for Deep Learning
by CodeSignal
Structured learning path that progresses from JAX NumPy basics and automatic differentiation through PyTrees, batching (vmap), and building real neural networks with Flax and Optax. Good for first contact with the ecosystem.
AI Distributed Training Basics — Train LLMs on Multiple GPU with JAX
by Udemy
Covers large-scale distributed training of GPT-style transformers with JAX and XLA, implementing techniques used by leading AI labs. Practical focus on multi-GPU and TPU training patterns.
Deep Learning with JAX and Flax
by Educative
Browser-based interactive course combining JAX fundamentals with the Flax neural network library, culminating in a hands-on deep learning project without requiring local GPU setup.
UvA Deep Learning Course — Introduction to JAX (Notebook Tutorial 2)
by Phillip Lippe (University of Amsterdam)
Free, academically rigorous Jupyter notebook that explains JAX's functional constraints, jaxpr intermediate representation, and GPU/TPU execution model. Part of a complete DL course translated from PyTorch to JAX+Flax.
📖 Books
Deep Learning with JAX
Grigory Sapunov · 2024
The primary and only dedicated book-length treatment of JAX for deep learning, published by Manning. Covers JAX primitives, differentiable model building, distributed computation with pmap, and the Flax ecosystem. Available on O'Reilly Learning Platform and Amazon.
🛠️ Tutorials & Guides
Getting Started with JAX for AI
The official entry-point tutorial from the JAX team. Assumes NumPy familiarity and walks through defining and training a neural network with Flax. Best first notebook after reading the JAX docs.
Flax NNX Basics
Official documentation for Flax NNX — the current recommended API for building neural networks in JAX. Covers the Module system, the functional API (split/merge/update), and integration with regular JAX code. Essential once JAX fundamentals are understood.
Get Started with JAX — Neural Network from Scratch (Tutorial 3)
Concise Colab notebook walking through building and training a neural network from scratch in raw JAX without a high-level library — ideal for understanding what Flax actually abstracts away.
Learning resources last updated: June 18, 2026