examples

Examples

A standalone pixi workspace that pulls kagglet from ../src and exposes one task per example. Each example is a directory with:

• notebook.yaml — kernel.name and sources are required; optional kernel.owner / kernel.title and other kernel settings. sources accepts explicit paths (e.g. [bootstrap.py, main.py]) or globs (e.g. ["*.py"], expanded against the example directory and sorted)
• one or more percent-format .py source files

Prereqs

1. Kaggle account.
2. OAuth login: KAGGLE_ENABLE_OAUTH=1 kaggle auth login.
3. Confirm the active Kaggle user: pixi run whoami.

Each example omits kernel.owner; the CLI fills it from your active Kaggle account. Set kernel.owner: alice only when explicitly pushing to a different owner.

Run

From this directory:

pixi run hello                  # push hello + poll until it finishes
pixi run hello-show             # dry-run: print derived kernel-metadata.json
pixi run gemma4-keras-tpu       # push gemma4-keras-tpu + poll
pixi run gemma4-keras-tpu-show  # dry-run for the gemma example

Each task is a thin wrapper over kagglet push <dir> --poll (or `kagglet show

`); see `pixi.toml`. Add a new example by dropping a directory next to the existing ones and registering one or two `[tasks.*]` entries.

Available examples

Dir	What it shows
hello/	Two percent-format .py sources auto-discovered into one notebook. No deps, CPU-only.
gemma4-keras-tpu/	Gemma 4 31B inference on Kaggle TPU (TpuV5E8) with resident JAX tensor sharding. Design notes and runtime env overrides live in the markdown cell at the top of main.py.