Skip to content

Welcome to distil labs

distil labs trains task-specific small language models (SLMs) from a prompt and a few dozen examples. A large teacher model does the expensive work: it judges whether your task is solvable, then generates the training data. A small student model learns to do that one task as well as the teacher can, from a model roughly 100x smaller and at 80%+ lower cost.

You supply the task and the examples. We handle the machine learning.

Getting started

Section titled “Getting started”

Launch Claude Code and give it this prompt:

Hey, pull the contents of distillabs.ai/onboarding.md
and get me onboarded from the instructions

Claude reads the onboarding script and works through it with you: installing the CLI, creating your account, picking a worked example, and training a model on it. It stops for your go-ahead at six points, so nothing runs that you haven’t seen.

Get started without Claude

Section titled “Get started without Claude”

Install the CLI and create an account, both from the terminal:

curl -fsSL https://cli-assets.distillabs.ai/install.sh | sh
distil signup      # already have an account? distil auth

distil signup opens your browser and hands the session back, so you finish signed in. Setup covers the rest: the Claude skill, and what to do if the installer doesn’t support your platform.

Minimal example

Section titled “Minimal example”
distil seed-dataset create --data ./my-data-dir
# Seed dataset created with ID: <seed-dataset-id>

distil training-dataset create-from-seed-dataset <seed-dataset-id>
# Training dataset created with ID: <training-dataset-id>

distil slm create-from-training-dataset <training-dataset-id>
# SLM created with ID: <slm-id>

distil deployment create-from-slm <slm-id>
distil deployment endpoint <deployment-id>     # URL and API key

That’s it. Your trained model is ready for hosted inference. Or you can run it locally.

Have production traces from an existing LLM application instead of a curated dataset? You can turn those into training data too.

Next steps

Section titled “Next steps”