AI Paper Atlas

Research map / Large language models / Alignment and preferences

AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback

Yann Dubois, Xuechen Li, Rohan Taori, Tianyi Zhang, Ishaan Gulrajani, Jimmy Ba, Carlos Guestrin, Percy Liang, Tatsunori Hashimoto

NeurIPS 2023 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 54 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Optimize a learned surrogate reward with PPO instead of stopping at SFTLLaMA 7B · AlpacaFarm evaluation, humanwin-rate against Davinci003: 44% to 55%Improved
Use more supervised fine-tuning data, SFT 52k instead of SFT 10kLLaMA 7B · AlpacaFarm evaluationwin-rate against Davinci003: 36.7\pm1.7 to 39.2\pm1.7 (simulated); 44.3\pm1.7 to 40.7\pm1.7 (human)No clear effect
Use Binary FeedME on preferred responses rather than SFT aloneLLaMA 7B · AlpacaFarm evaluationwin-rate against Davinci003: 36.7\pm1.7 -> 36.6\pm1.7 (simulated); 44.3\pm1.7 -> 37.9\pm1.7 (human)No clear effect
Use binary reward conditioning on pairwise preferences rather than SFT aloneLLaMA 7B · AlpacaFarm evaluation, simulatedwin-rate against Davinci003: 36.7\pm1.7 -> 32.4\pm1.6Hurt
Use PPO rather than Expert Iteration to optimize the surrogate rewardLLaMA 7B · AlpacaFarm evaluationwin-rate against Davinci003: 41.9\pm1.7 -> 46.8\pm1.8 (simulated); 45.7\pm1.7 -> 55.1\pm1.7 (human)Improved
Increase best-of-n sampling from the SFT modelLLaMA 7B · AlpacaFarm evaluationwin-rate against Davinci003: 36.7\pm1.7 -> 45.0\pm1.7 (simulated); 44.3\pm1.7 -> 50.7\pm1.8 (human)Improved
Train with the top reward bin and KL regularization in QuarkLLaMA 7B · AlpacaFarm evaluation, simulatedwin-rate against Davinci003: 36.7\pm1.7 -> 35.6\pm1.7Hurt
Add label-flip noise to simulated preference labels (ablation)AlpacaFarm training annotator · Best-of-n reward optimizationsimulated win-rate as sample count increases: Label-flipping ratios tested: 0\%, 17.5\%, 25\%, 35\%Mixed
Use a pool of 13 annotators instead of a single deterministic GPT-4 prompt (ablation)AlpacaFarm preference simulator · Reward-model over-optimizationwin-rate as reward optimization increasesImproved
Use label noise and a pool of annotators to reproduce human variability (ablation)AlpacaFarm preference simulator · Reward-model over-optimizationwin-rate as reward optimization increasesImproved
Use a single low-variance GPT-4 annotator rather than the variable AlpacaFarm annotator pool for simulated PPO training (ablation)PPO, LLaMA 7B · Human evaluationwin-rate: 43% -> 50%Improved
Normalize PPO advantages across the full rollout batch rather than separately for each minibatch (ablation)PPO implementation · PPO training stabilitytraining stabilityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy