Skip to content

Getting started

Everything you need to go from a new account to a fine-tuned model of your own, in three short steps.

What Tuneline Studio is

Tuneline Studio is a web app for fine-tuning open large language models without writing training code. You build a pipeline on a canvas by connecting blocks: load a dataset, format it as chat, pick a base model, train it with SFT, DPO or GRPO, evaluate it against the base model, check it with a quality gate, then merge it, export it to GGUF, publish it to the Hugging Face Hub or serve it from an endpoint. Every block has a settings form, so you can change any step without touching Python.

When the pipeline is ready, you choose where it runs: a free Kaggle GPU, a notebook in Google Colab or any Jupyter server, Modal, RunPod, Hugging Face Jobs, or your own computer. You then watch the run live, with each step's status, the logs, loss curves, evaluation reports and links to the finished model. Tuneline uses your own provider keys and accounts: providers bill you directly, and your models go to your own Hugging Face account.

Who it is for

  • ML practitioners who want repeatable fine-tuning runs, side-by-side evaluations and a model registry without writing glue code for each GPU provider.
  • Analysts and domain experts, for example compliance teams, who have labeled examples and want a small model that makes the same decisions, with a quality gate that blocks a model that isn't good enough.
  • Students and hobbyists who want to learn fine-tuning on free GPUs, or on a laptop CPU with the smoke test.

Your path in three steps

  1. Sign up. Create an account in the hosted studio. Nothing needs to be installed. See Install.
  2. Make your first run. Run the CPU smoke test (no GPU needed) template. It trains a tiny model for 20 steps and evaluates it in a few minutes, which checks your whole setup. See Your first pipeline in 10 minutes.
  3. Train your own model. Use the Sanctions alert triage template to fine-tune a model that decides whether a screening alert should be escalated, publish it to your Hugging Face account and use it in a small app. See Fine-tune and use your own model.

The Pipelines page before you create your first pipeline

Pages in this section

Page What you learn
Install Create an account, connect your own computer (optional), or host your own copy.
Your first pipeline in 10 minutes Build and run the CPU smoke test and read its results.
Fine-tune and use your own model Train, gate and publish a real model, then call it from an app.
Try Tuneline online Which compute options are free, which are paid, and a no-GPU path.

What you need

You need Why Notes
A web browser Tuneline Studio runs in the browser at https://tuneline-studio.vercel.app. Works best on a screen at least 1280 px wide.
A Hugging Face account and a write token Almost every pipeline downloads models from the Hub and pushes results (adapters, merged models, cached steps) to your account. Free. Create the token at huggingface.co → Settings → Access Tokens, then add it in Tuneline under Settings → API keys. The CPU smoke test runs without it.
Somewhere to run Training needs a machine, ideally with an NVIDIA GPU. Free options: Kaggle (free T4 GPUs, phone-verified account), Notebook (Google Colab, Lightning or any Jupyter), My computer (your own PC, GPU or CPU), and Modal (monthly free credit on its Starter plan). See Try Tuneline online.

Tip

After you first sign in, the Pipelines page shows a short checklist: add a Hugging Face token, pick where to run, and try the CPU smoke test. Following it is the quickest way to a working setup.

Where to go next