---
title: "Data Preparation"
description: "Follow one JSON example into tokens: who speaks, which answer belongs to which input, which tokens the loss sees, and why related examples stay together."
canonical_url: "https://fanout.sh/ai/lessons/fine-tuning-data-preparation"
md_url: "https://fanout.sh/ai/lessons/fine-tuning-data-preparation.md"
access: "pro-preview"
---

# Data Preparation

Follow one JSON example into tokens: who speaks, which answer belongs to which input, which tokens the loss sees, and why related examples stay together.

- Access: Fanout Pro (public preview)
- Module: How to Fine-Tune Models

## A glimpse of Data Preparation

An instruction-following model learns from the responses you mark as desirable. A neat JSON file is only the container. The training example must also say who is speaking, what answer belongs to that input, and which tokens should contribute to the loss.

Consider a small routing task. The user writes “package late”; the desired assistant answer is delivery.

## Continue learning

Pro unlocks guided learning, complete notes, practical walkthroughs, exercises, and every AI Research lesson.

---
This representation contains public Fanout content only. Protected Pro lessons, account data, billing, checkout, and pricing are not included.

Browse the public content map: https://fanout.sh/sitemap.md