All research briefings
2023
Alignment
8 min read

DPO

Direct Preference Optimization: Your Language Model is Secretly a Reward Model

Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, Chelsea Finn

Editorially reviewed July 27, 2026

The paper in one sentence

DPO converts preference alignment into a supervised-style classification objective over chosen and rejected responses.

What the paper does

DPO derives a direct optimization objective for preference data. Instead of fitting a separate reward model and then running reinforcement learning, it trains a policy to increase the relative likelihood of preferred responses while staying anchored to a reference policy.

Why it matters to engineers

The method simplifies the operational pipeline for preference tuning. It removes a separately deployed reward-model optimization loop, though data quality, reference-model choice, evaluation, and safety controls remain decisive.

Key claims from the paper

  • The optimal policy for the regularized reward objective can be expressed in closed form through the policy and reference policy.
  • Preference optimization can therefore be performed with a simple binary cross-entropy objective.
  • The evaluated method matches or exceeds the paper’s PPO-based RLHF baselines on several tasks while being simpler to train.

Architecture, step by step

  1. 1

    Preference pairs

    Each example contains a prompt, a preferred response, and a rejected response.

  2. 2

    Reference policy

    A frozen model provides the anchor used to regularize how far the trained policy moves.

  3. 3

    Policy scoring

    The trainable model assigns likelihoods to both responses.

  4. 4

    Direct loss

    A classification-style objective increases the relative margin for the preferred response.

Production takeaways

  • Preference-data construction and annotator agreement should be monitored like product metrics, not treated as a static dataset detail.
  • Track capability, safety, style, and distribution-shift evaluations separately after tuning.
  • The reference model and regularization strength control how aggressively behavior changes.
  • Maintain lineage from each tuned checkpoint to its preference dataset, policy version, and evaluation suite.

Limitations and caveats

  • The method inherits bias and ambiguity from pairwise preference data.
  • Offline preferences may not cover adversarial or long-horizon behavior.
  • A simpler training objective does not remove the need for safety evaluation or online monitoring.

Interview questions to test understanding

  1. What pipeline stages does DPO remove compared with a classic reward-model-plus-PPO workflow?
  2. Why is a reference policy used in the DPO objective?
  3. How would you detect preference-data shortcuts or annotator bias?
  4. Which evaluations would you require before deploying a preference-tuned model?
Browse all research