# DPO / KTO Dataset Engineering: The Engineering of Chosen/Rejected Triplet Generation

> Source: https://sukruyusufkaya.com/en/learn/fine-tuning-cookbook/ftc-dpo-kto-dataset-engineering
> Updated: 2026-05-14T14:42:51.155Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part II — Tokenizer & Data Engineering
**TLDR:** DPO and KTO need 'chosen' (good) and 'rejected' (bad) response pairs. Generation methods: AI Feedback Loop (RLAIF), regex-graded pairs (math/code), human-in-the-loop, hard-negative mining, length-controlled pairs. UltraFeedback analysis, TR DPO dataset build, KTO's unpaired advantage.