English
← Back to AI Technology

Training and post-training

SFT / RLHF / DPO

A family of post-training methods for teaching instruction following and preference-aligned behavior. SFT learns from demonstrations; RLHF and DPO use preference comparisons through different optimization procedures.

Reference period
2017 / 2022 / 2023
Last reviewed

Key terms

  • supervised fine-tuning
  • preference data
  • reward model
  • alignment

Connected across the project

Builds on

Used by

Related technology

Related papers

Historical context

Primary sources

  1. Deep reinforcement learning from human preferences
  2. Training language models to follow instructions with human feedback
  3. Direct Preference Optimization

This is a curated technical map, not a claim of comprehensive coverage.