ManySpace
Türkçe Sign in Sign up

AI & MLRecommendedAIML-LM-025

Direct Preference Optimization

Skipping the explicit reward model: a classification loss on preference pairs derived from the KL-regularised objective.

Difficulty
4 / 5
Importance
4 / 5
Study time
1–2 h
Kind of work
Mixed

Why it matters

The learner can compute the DPO loss for one preference pair and explain how the implicit reward relates to the policy/reference log-ratio.

The route to this topic

Route AIML-LM-025
What Is a Model? Tasks: Regression, Classification, and Beyond Features and Targets Classification Framing Linear Regression I: Prediction and Residuals Embeddings Tokens and Tokenization Scores and Logits Loss and Fitting: Squared Error Sigmoid and Probabilities Multiclass Scores and Softmax The Artificial Neuron Autoregression Objective Functions and Empirical Risk Logistic Regression Activation Functions Next-Token Prediction Gradient Descent in Machine Learning Likelihood The Multilayer Perceptron Binary Cross-Entropy Cross-Entropy Loss Stochastic and Mini-batch Optimization Transfer Learning The Language-Modeling Objective Fine-Tuning Pretraining Supervised Fine-Tuning Preference Optimization Direct Preference Optimization