Supported models

Which base models the Arkor backend accepts today, why the list is short, and where it’s headed.

Today both the model field of createTrainer and Studio’s Playground (base-model mode) accept exactly one value: gemma-4-E4B-it. Below: what that value gets you, why the list is short on purpose, and what’s next.

What you can pick today

gemma-4-E4B-it is the Gemma 4 instruction-tuned build, packaged by Unsloth for fast LoRA / QLoRA fine-tuning. Every starter template (triage, translate, redaction) targets it by default.

import { createTrainer } from "arkor";

export const trainer = createTrainer({
  name: "support-bot-v1",
  model: "unsloth/gemma-4-E4B-it",
  dataset: { type: "huggingface", name: "arkorlab/triage-demo" },
});

Why one model, right now

The priority in alpha is making both training and inference feel fast on one base model, not building a wide catalog. Tuning the whole path from arkor dev through arkor start to a chat-ready checkpoint against one model means optimizations to that model compound across runs, instead of getting diluted across many.

Gemma 4 fits this stage because it is small enough to iterate on quickly, capable enough to be useful inside a product, and openly licensed.

What’s coming

Gemma 4 family

Open the model field to the full Gemma 4 family so you can pick the variant that matches your use case (size, capability, latency, quality).

Other open-weight families

Expanding to additional open-weight families is on the roadmap Backlog. Today the focus stays on Gemma 4 to keep delivery on it fast and stable.

See also

• Trainer concept for what the model field is and where it sits among the other fields.
• createTrainer reference for the full TrainerInput type.
• Roadmap for the broader picture of what’s next.