Recipes

Recipes that show what TypeScript unlocks for an Arkor training run today.

Arkor is alpha. The framework’s surface is intentionally small: one trainer, one manifest slot, a handful of lifecycle callbacks. What makes that small surface worthwhile is that all of it is your TypeScript code. Anything you can express in TS, you can wire into a training run, with the same editor, types, and review flow as the rest of your product.

This section collects the patterns that show up first when you try. Each recipe stays inside today’s public SDK: no roadmap APIs, no internal imports, nothing that requires forking the runtime.

Recipes

Recipe	What it shows	Built on
Mid-run evaluation	Sanity-check the half-trained model against a fixed prompt at every checkpoint, before the run finishes.	onCheckpoint({ infer })
Early stopping on diverging loss	Abort a run automatically when the loss curve goes the wrong way, and stop the GPU on the backend too.	onLog, AbortSignal, trainer.cancel()
Slack / Discord notifications	Post to a webhook on completion or failure, without leaving the trainer file.	onCompleted / onFailed, fetch
Programmatic runs (no CLI)	Drive training from a Next.js API route, a cron worker, or CI without going through arkor dev / arkor start.	runTrainer, Trainer.start / wait
Customizing the starter templates	Treat the scaffolded templates as starting points. Change the dataset, hyperparameters, callbacks, and base model.	createTrainer, DatasetSource

A note on callback exceptions before you start

Three of the recipes below put logic inside the lifecycle callbacks. The runtime catches errors thrown out of a callback and routes them through the SSE reconnect loop (SDK § Lifecycle callbacks), and maxReconnectAttempts defaults to unlimited. In practice that means a throw inside a callback can be silently retried.

The recipes here use a simple convention to stay deterministic:

• State changes go through outer variables. Use an AbortController, a closure flag, or a returned Promise rather than throwing.
• Side effects are guarded with try / catch inside the callback. If a Slack post fails, log it and continue; do not let it bubble.

This is a pattern, not a limit on what you can do. Once you have it, the recipes compose cleanly.

Things this section does not cover

• Production deploys. Arkor today only runs on managed GPUs, and createArkor’s deploy slot is a reserved type field with no implementation. Serving recipes will land when that surface lands.
• Multiple trainers per project. createArkor accepts a single trainer; running several together is a programmatic-run pattern (see the recipe), not a manifest pattern.
• Custom base models beyond what the backend accepts. The model field is forwarded to the cloud API verbatim; today the curated path is Gemma. Recipes do not pretend other models work end to end.