# Examples ```{toctree} :hidden: examples/fashion_mnist examples/simple_demo examples/example_async_jobs examples/example_data_api examples/example_checkpoint examples/example_keras_checkpoint examples/example_collections examples/example_gke examples/pathways_example examples/gemma_sft_pathways_distributed examples/gemma3_sft_demo examples/tunix_sft ``` A catalog of runnable example scripts using Kinetic. Every example below is rendered directly on this site and is also available as a raw Python script in the GitHub repository. Tier badges: - **Quickstart:** your first run. Minimal setup, sensible defaults. - **Core:** the everyday product surface: async jobs, data, checkpoints, parallel sweeps. - **Advanced:** multi-host Pathways jobs, LLM fine-tuning, anything that needs special quota or external credentials. To run any example: clone the repo, install Kinetic, set `KINETIC_PROJECT`, and `python examples/.py`. ```bash git clone https://github.com/keras-team/kinetic.git cd kinetic uv pip install -e . export KINETIC_PROJECT="your-project-id" python examples/fashion_mnist.py ``` ## Quickstart ::::{grid} 1 2 2 3 :gutter: 3 :class-container: sd-text-left :::{grid-item-card} Fashion-MNIST on a TPU :link: examples/fashion_mnist.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 The first thing to run after `kinetic up`. A small Keras classifier on Fashion-MNIST that confirms your cluster can schedule a TPU pod and stream a real result back to your shell. +++ {bdg-secondary}`Keras`   {bdg-secondary}`TPU` ::: :::{grid-item-card} Keras + JAX smoke test :link: examples/simple_demo.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 The cheapest sanity check there is. Keras-on-JAX on a CPU node — no accelerator quota needed, useful for verifying your install before you ask for hardware. +++ {bdg-secondary}`Keras`   {bdg-secondary}`JAX`   {bdg-secondary}`CPU` ::: :::: ## Core ::::{grid} 1 2 2 3 :gutter: 3 :class-container: sd-text-left :::{grid-item-card} Submit, monitor, and reattach :link: examples/example_async_jobs.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 Walks through every part of the detached-job API end-to-end: `run_async()`, `status()`/`tail()`/`result()`, reattach from another shell with `kinetic.attach()`, and enumerate jobs with `list_jobs()`. +++ {bdg-secondary}`Async`   {bdg-secondary}`Reattach` ::: :::{grid-item-card} Ship local files into the job :link: examples/example_data_api.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 Wrap a local directory in `kinetic.Data(...)` and let it land as a plain filesystem path on the remote — your training code doesn't have to know whether the bytes started on your laptop or in GCS. +++ {bdg-secondary}`Data`   {bdg-secondary}`GCS` ::: :::{grid-item-card} Resumable JAX training with Orbax :link: examples/example_checkpoint.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 JAX training that picks up where it left off. Writes Orbax checkpoints to `KINETIC_OUTPUT_DIR` and proves the resume path by relaunching the same function and seeing it skip already-completed steps. +++ {bdg-secondary}`JAX`   {bdg-secondary}`Checkpointing`   {bdg-secondary}`Orbax` ::: :::{grid-item-card} Resumable Keras training :link: examples/example_keras_checkpoint.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 Auto-resumable Keras training. Round-trips `model.get_weights()` through Orbax so a restarted job picks up at the right step without any custom save/load code. +++ {bdg-secondary}`Keras`   {bdg-secondary}`Checkpointing`   {bdg-secondary}`Orbax` ::: :::{grid-item-card} Parallel hyperparameter sweep :link: examples/example_collections.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 Fan out a grid of jobs with `run_async_map()`, batch submissions to keep the cluster happy, and gather results — including how to handle the job that inevitably fails halfway through. +++ {bdg-secondary}`Sweep`   {bdg-secondary}`Parallel` ::: :::{grid-item-card} Mix accelerators in one driver :link: examples/example_gke.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 One driver script that successively schedules work on CPU, TPU, and GPU pools — handy for verifying which hardware your cluster will actually serve. +++ {bdg-secondary}`Multi-accelerator`   {bdg-secondary}`Cluster` ::: :::: ## Advanced ::::{grid} 1 2 2 3 :gutter: 3 :class-container: sd-text-left :::{grid-item-card} Multi-host JAX on Pathways :link: examples/pathways_example.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 The reference for scaling beyond a single TPU host. A short JAX program that verifies cross-host collectives are actually wired up before you trust them with a real workload. +++ {bdg-secondary}`JAX`   {bdg-secondary}`Pathways`   {bdg-secondary}`Distributed` ::: :::{grid-item-card} Distributed Gemma 2B fine-tune :link: examples/gemma_sft_pathways_distributed.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 End-to-end SFT of Gemma 2B with LoRA across multiple TPU hosts. The realistic LLM workload to model your own fine-tuning runs after — pulls weights from Kaggle and runs on Pathways. +++ {bdg-secondary}`LLM`   {bdg-secondary}`Pathways`   {bdg-secondary}`Distributed` ::: :::{grid-item-card} Single-TPU Gemma 3 fine-tune :link: examples/gemma3_sft_demo.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 Compact Gemma 3 1B SFT on a single TPU. A good baseline for getting an LLM workload running before scaling out to Pathways, and a worked example of forwarding Kaggle credentials into the remote pod. +++ {bdg-secondary}`LLM`   {bdg-secondary}`TPU` ::: :::{grid-item-card} Tunix SFT Example :link: examples/tunix_sft.md :class-card: sd-shadow-sm :class-body: sd-fs-6 :class-title: sd-fs-5 SFT of Gemma 3 with LoRA/QLoRA on TPU v5litepod. Demonstrates how to run the Tunix SFT script on a remote cluster with environment variable capture for credentials. +++ {bdg-secondary}`LLM`   {bdg-secondary}`TPU`   {bdg-secondary}`LoRA` ::: :::: ## Related pages - [Getting Started](getting_started.md): your first run, end-to-end. - [Keras Training](examples/keras_training.md): patterns for Keras users. - [LLM Fine-tuning](examples/llm_finetuning.md): extended walkthrough using the Gemma examples.