What would you
like to build?

Train neural networks

Train a CPU image classifier, preserve its complete state through interruption, inspect held-out errors and verify its exported inference. Language modeling and TPU work are optional branches.

Available lesson drafts · 53 lessons

First artifact: A CPU classifier with stable loss, replay checks and count-weighted evaluation.

19 available in the focus phases. Total above includes preparation.

Project available: A recoverable image classifier with verified exported inference

Reproducible JAX/Flax experiments workshop

Optional TPU and Google Cloud specialization

Optional language-model branch

Optional staged embedding-model pilot

Learn by interacting

Train a policy from actual environment interaction, verify its objective against an independent small-state reference, and compare multiple seeded agents on fixed held-out evaluation streams.

Available lesson drafts · 38 lessons

First artifact: A tested environment transition table and reproducible batched trajectories.

4 available in the focus phases. Total above includes preparation.

Project available: Train and audit a policy from interaction

Optional TPU and Google Cloud specialization

Learn TPUs and Google Cloud

Build and recover an experiment on CPU, then learn GCP provisioning, TPU generations, training and inference memory, precision policies and XProf profiling. Connect measured error and real target traces to model engineering.

Available lesson drafts · 87 lessons

First artifact: A verified TPU runtime receipt, a checkpoint-resumed TPU experiment log, and a memory/precision/XProf decision dossier.

35 available in the focus phases. Total above includes preparation.

Project available: Causal Transformer training-system capstone

Open the TPU & GCP practical guide

TPU generations, precision and profiling

Ship and adapt JAX workloads

Move trained computations into a measured inference service. Check framework/precision parity, package a non-root Docker image, verify readiness and connect evaluation, model identity and rollback evidence.

Available lesson drafts · 53 lessons

First artifact: An exported computation checked against its source on the same inputs.

8 available in the focus phases. Total above includes preparation.

Project available: Verify an inference artifact and capacity hypothesis

Optional TPU and Google Cloud specialization

Simulate and discover

Simulate physical dynamics, check numerical and derivative accuracy, and recover unknown parameters with held-out evidence. Build a reusable CPU inverse-problem audit before extending to adaptive solvers or accelerators.

Available lesson drafts · 33 lessons

First artifact: A cooling simulator with a convergence table and an independent sensitivity check.

4 available in the focus phases. Total above includes preparation.

Project available: Audit a differentiable scientific inverse problem

Model uncertainty

Express assumptions as probability models, verify a known posterior, then implement HMC and variational inference with explicit diagnostics. Use predictive checks to distinguish a sampler problem from a model that does not explain the data.

Available lesson drafts · 33 lessons

First artifact: An exact Bayesian regression posterior checked against an independent host calculation.

4 available in the focus phases. Total above includes preparation.

Project available: Audit a Bayesian regression model

Understand JAX itself

Trace a function into jaxpr, verify forward and reverse sensitivities, write exact custom derivative rules, and compile a small transformation whose limits you can explain.

Available lesson drafts · 33 lessons

First artifact: An annotated jaxpr that explains a computation and a changed input shape.

4 available in the focus phases. Total above includes preparation.

Project available: Audit derivatives and a tiny transformation

Make it fast. Make it scale.

Profile a correct workload, preserve its global objective while partitioning arrays, and audit actual Pallas kernels before requesting target measurements. CPU semantics are available; accelerator speed and multi-host claims require their own receipts.

Available lesson drafts · 53 lessons

First artifact: A synchronized timing report and a checked array placement on logical CPU devices.

13 available in the focus phases. Total above includes preparation.

Project available: A checked kernel and a measurement-backed optimization argument

Optional TPU and Google Cloud specialization

Operate JAX workloads

Operate reproducible model workloads from tracked experiments to containerized inference. Practice failure recovery, MLOps release gates, LLMOps evaluation and tracing, and ModelOps ownership, approval and rollback.

Available lesson drafts · 65 lessons

First artifact: An interrupted CPU run whose next data batches, losses and full state match uninterrupted execution.

31 available in the focus phases. Total above includes preparation.

Project available: An observable, restartable local workload

Reproducible JAX/Flax experiments workshop

Optional TPU and Google Cloud specialization