A ready-to-use workspace
Follow the setup lesson to choose an interpreter, install the course environment and save your first experiment.
Open setup lesson →Download beginner workspaceHomeTake the course with you
Practice on your CPU, keep your own evidence, and learn in the browser, a notebook or your agent.
Run something real
Each download includes the work to try and the checks to explain. Reference execution and your own exercise evidence stay separate.
Follow the setup lesson to choose an interpreter, install the course environment and save your first experiment.
Open setup lesson →Download beginner workspaceInspect local shards, compare replication with partitioning, and repair an uneven batch on one physical host.
Open virtual-device lesson →Download notebookRestart the notebook kernel, then Run all. Logical CPU devices do not emulate TPU performance.
Build preprocessing, check gradients, fit a model and produce a report through four inspectable stages.
Open project →Download starter, checks & referenceConnect model shapes, training updates and evaluation checks in a small synthetic classification experiment.
Open classifier project →Download project bundleAway from the browser
The book includes available lesson drafts, their worked code, experiments, practice and references. Unwritten roadmap topics are excluded.
The filename retains “foundations”; both editions contain all available lesson drafts from the current build.
Your agent host
Ask for a starting point, work through an exercise and keep a study plan in LEARNING.md. Skills need local course files; they do not install the course or its Python environment. Download the course workspace for the current lessons and tutors together.
Read the skill files first. With Node.js/npm available, run this in the extracted course folder or repository root and choose your agent when prompted:
npx skills add . --skill start-learning learn-jax jax-course-guide check-jaxProject installation is the default. To select Codex explicitly, add --agent codex. No global installation is required.
The current tutors are not yet published on GitHub. After a revision containing these skills is published:
npx skills add mlnomadpy/jaxpathways --skill start-learning learn-jax jax-course-guide check-jaxA private repository requires configured Git/GitHub access. Use the local command above for the current working source. The remote command becomes available once these skills are published.
Without Node.js, download the skill bundle and copy the desired skill folders into the directory your agent host supports. The bundle also includes the contributor skill create-jax-course.
Skills CLI installation documentation ↗ · Repository skill instructions ↗
Keep the extracted course folder or checkout open. In Codex, choose start-learning or type $start-learning. In Claude Code, use /start-learning. In another host, say “Use start-learning to help me learn JAX from this workspace.”
The tutor keeps your goal, next lesson, attempts and review queue in LEARNING.md. Use learn-jax next time to continue. Browser-local progress is separate; neither store synchronizes automatically.
From your course checkout
Browse availability, read a lesson or run its reference companion. Execution needs the tested packages in requirements-cpu.txt.
Write your own exercise solution in the lesson script or notebook first, then run the verifier to check your assertions.
python3 scripts/course.py list --available
python3 scripts/course.py list --pathway ship
python3 scripts/course.py show first-gradient
python3 scripts/course.py run first-gradient
python3 scripts/course.py quiz first-gradient
python3 scripts/course.py plan inference-engineerThe final command prints a proposed route. Keep the selected next lesson and your own evidence in your learning plan.
For course authors and integrations
Edit curriculum/course.json for membership and ordering, learning-paths/*.json for routes, and the canonical lesson.json for teaching material. Build the generated documents, notebooks, website resources and books from that source.
npm run build
npm run check
npm run smoke:cpuUse create-jax-course to guide authoring, run npm run smoke:cpu to record CPU execution receipts, and stage accelerator benchmarks on a Cloud TPU VM.
These are static JSON resources, with authored/planned availability. They provide no remote execution or progress service.