Skip to content
ProfitOps · PFN Studio PFN Studio

Search is only available in production builds. Try building and previewing the site to test it out locally.

Getting started

This walkthrough takes you from a fresh sign-up to a trained brain that predicts on your own data. No GPU required.

1. Sign up

Go to cloud.pfnstudio.com/register and create an account. After confirming email you’ll land on the Welcome page (/welcome), where you pick a persona:

  • Walk me through it (🎓 Basic mode) — the wizard guides every choice, plain-English labels throughout
  • Give me the tools (🛠️ Developer mode) — direct access to projects, priors, models, evals, runs, YAML

Pick Basic for this walkthrough — you can switch later from the avatar menu (see Basic vs Developer mode).

2. Open the wizard

Basic mode lands on /brains (your brain index). Click 🧠 Teach a brain in the top nav.

You’ll see a 5-step wizard:

  1. Capabilities — what your brain should be able to do
  2. Speciality — auto-filled grouping (usually no change)
  3. Brain — model architecture (auto-filled recipe size)
  4. Goal — benchmarks the brain should pass (auto-filled defaults)
  5. Practise — training length + compute target

The capability picker on step 1 auto-fills steps 2–5. Most users hit ⚡ Train now on step 1 without touching the others. See The wizard for the full step-by-step.

3. Pick a capability

For your first brain, pick 🔢 Predict a number — the simplest live capability, backed by the Müller 2022 PFN reference paper.

The “Train now” button appears as soon as you tick a capability. The footer at the bottom shows you’ll train a Steady brain on Bayesian regression practice data for Standard practise (~5 minutes) on your laptop (CPU).

4. Train

Click ⚡ Train now. The wizard creates a project and dispatches a training run. You land on /brain/<id> — your brain’s dashboard.

While training:

  • A progress bar shows steps + a live loss curve
  • A live log streams short status messages (“Brain is settling into the pattern…”)
  • You can ⏹ Cancel practise at any time

When training finishes, the page transitions to the Ready state. The hero now says “Your brain can predict a number”. You’ll see a What this brain can do card with the capabilities it learned, and a How well it learnt card with goal evals.

5. Try it on real data

The Try it on your data section gives you two textareas:

  • Examples (with answers) — a small reference set the brain learns from in-context
  • Questions (predict for these) — rows you want predicted

For a regression brain, paste one row per line with x, y (2 columns) on the example side, then just x on the question side. Click ✨ or use a sample to see how it works to auto-fill with synthetic data — useful the first time.

Hit Predict and the predictions appear in a table.

6. Where to go next

  • Try more capabilities — pick Show confidence + Forecast forward in a new brain to see how mixing works. Capabilities reference.
  • Switch to Marathon (GPU) training — for a paper-scale brain. Needs a Vast.ai GPU rental.
  • Browse the marketplace — install priors and brains others have published. Marketplace.
  • Switch to Developer mode — full control over priors, models, evals, runs. Developer mode.
  • Learn the vocabulary — Glossary.

What you don’t need

  • No GPU for Quick / Standard / Long practise — runs on your laptop CPU
  • No PyTorch install on your laptop — training runs on PFN Studio’s hosted workers (or Vast.ai for Marathon)
  • No CLI for the basic flow — everything is in the browser. The CLI is for advanced sync workflows (see API tokens)