Skip to content

Try Tuneline online

You can try Tuneline Studio without paying anything: sign up in the browser, then run pipelines on free compute such as Kaggle's GPUs, a Colab notebook or your own computer.

Use the hosted studio

Open https://tuneline-studio.vercel.app and create an account (see Install). Tuneline doesn't charge for compute: every run uses your own provider accounts and keys, and providers bill you directly. Building pipelines, browsing the model catalog, using the hardware advisor and uploading datasets need no compute at all.

Free places to run

Where to run What you get What you need
Kaggle (free) A notebook with two T4 GPUs (16 GB each); about 30 GPU hours a week. Slower than paid GPUs, but costs nothing. A Kaggle account verified with a phone number, and a Kaggle API token in Settings → API keys.
Notebook Download a notebook and run it in Google Colab, Lightning AI or any Jupyter server. Colab's free tier offers a GPU when one is available; its limits vary. Nothing in Tuneline. An account with the notebook service.
My computer Your own NVIDIA GPU, or the CPU for the smoke test. A one-time setup in Settings → Your computers. See Install.
Modal Serverless GPUs billed per second. The Starter plan includes $30 of compute a month at the time of writing, which covers many small runs. A Modal token ID and secret in Settings → API keys.

Free tiers and quotas are set by each provider and change over time. Check the provider's own page before you rely on them.

Where to run How it is billed What you need
RunPod GPU pods billed per second; the pod stops when the run ends. A RunPod API key.
Hugging Face Jobs Billed to your Hugging Face account for the time the job runs. Your Hugging Face token, and billing (credit) on your Hugging Face account.
Modal, beyond the included credit Per second. A Modal token ID and secret.

Before every run, the Run pipeline dialog shows the estimated GPU memory, time and cost for each GPU, and lets you set Stop after (hours) and Stop at cost (USD) so a run can't run away with your budget. For current prices across providers, open the Hardware page in the app and select Refresh live prices. See Hardware and costs.

Serving a model is also paid. Deploy endpoint and the Deployments page start OpenAI-compatible endpoints on RunPod Serverless, Hugging Face Inference Endpoints or Modal. They can scale to zero when idle, but you pay while they run. See Deployments.

What's free on Hugging Face

Almost every pipeline uses a Hugging Face account, so it helps to know which parts cost money.

Free with a Hugging Face account Paid
Creating an account and access tokens. Hugging Face Jobs (training runs on Hugging Face hardware).
Downloading open models and datasets, including gated ones once you accept their license. Inference Endpoints (serving a model on dedicated hardware).
Storing your fine-tuned models, adapters and cached steps in private repositories, within Hugging Face's storage limits for free accounts. Inference Providers usage beyond any credits your plan includes (used by the Playground to chat with Hub models, and by the LLM judge).

Check Hugging Face's pricing page for current prices, credits and storage limits.

Try it without a GPU

You don't need a GPU to see a whole run from start to finish:

  1. Create a pipeline from the CPU smoke test (no GPU needed) template.
  2. Run it on My computer (any computer works) or as a Notebook in Colab with the default CPU runtime.
  3. Watch it train a 135M-parameter model for 20 steps and evaluate it, in about a minute on a typical CPU once the model has downloaded.

The step-by-step tutorial is Your first pipeline in 10 minutes.

Without running anything, you can also:

  • browse the Models page for models you can fine-tune, with their license and memory needs;
  • use the Hardware page to see which GPUs fit a model and what a run would cost;
  • open any template and look at its steps and settings.

Next steps