Authors:
(1) Rafael Rafailo, Stanford University and Equal contribution; more junior authors listed earlier;
(2) Archit Sharma, Stanford University and Equal contribution; more junior authors listed earlier;
(3) Eric Mitchel, Stanford University and Equal contribution; more junior authors listed earlier;
(4) Stefano Ermon, CZ Biohub;
(5) Christopher D. Manning, Stanford University;
(6) Chelsea Finn, Stanford University.
Table of Links
4 Direct Preference Optimization
7 Discussion, Acknowledgements, and References
A Mathematical Derivations
A.1 Deriving the Optimum of the KL-Constrained Reward Maximization Objective
A.2 Deriving the DPO Objective Under the Bradley-Terry Model
A.3 Deriving the DPO Objective Under the Plackett-Luce Model
A.4 Deriving the Gradient of the DPO Objective and A.5 Proof of Lemma 1 and 2
B DPO Implementation Details and Hyperparameters
C Further Details on the Experimental Set-Up and C.1 IMDb Sentiment Experiment and Baseline Details
C.2 GPT-4 prompts for computing summarization and dialogue win rates
D Additional Empirical Results
D.1 Performance of Best of N baseline for Various N and D.2 Sample Responses and GPT-4 Judgments
C.2 GPT-4 prompts for computing summarization and dialogue win rates
A key component of our experimental setup is GPT-4 win rate judgments. In this section, we include the prompts used to generate win rates for the summarization and dialogue experiments. We use gpt-4-0314 for all our experiments. The order of summaries or responses are randomly chosen for every evaluation.
Summarization GPT-4 win rate prompt (S).
Which of the following summaries does a better job of summarizing the most \ important points in the given forum post?
Post:
Summary A:
Summary B:
FIRST provide a one-sentence comparison of the two summaries, explaining which \ you prefer and why. SECOND, on a new line, state only "A" or "B" to indicate your \ choice. Your response should use the format: Comparison: Preferred: <"A" or "B">
Summarization GPT-4 win rate prompt (C).
Which of the following summaries does a better job of summarizing the most \ important points in the given forum post, without including unimportant or \ irrelevant details? A good summary is both precise and concise.
Post:
<post>
Summary A:
<Summary A>
Summary B:
<Summary B>
FIRST provide a one-sentence comparison of the two summaries, explaining which \ you prefer and why. SECOND, on a new line, state only "A" or "B" to indicate your \ choice. Your response should use the format:
Comparison: <once-sentence comparison and explanation>
Preferred: <"A" or "B">
Dialogue GPT-4 win rate prompt.
For the following query to a chatbot, which response is more helpful?
Query: <the user query>
Response A:
<either the test method or baseline>
Response B:
<the other response>
FIRST provide a one-sentence comparison of the two responses and explain \ which you feel is more helpful. SECOND, on a new line, state only “A“ or \ “B“ to indicate which response is more helpful. Your response should use\ the format:
Comparison: <one-sentence comparison and explanation>
More helpful: <“A“ or “B“>
This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license.