ORPO: Monolithic Preference Optimization without Reference Model
Paper
•
2403.07691
•
Published
•
59
RLHF
-
-
sDPO: Don't Use Your Data All at Once
Paper • 2403.19270 • Published • 32 -
Teaching Large Language Models to Reason with Reinforcement Learning
Paper • 2403.04642 • Published • 46 -
Best Practices and Lessons Learned on Synthetic Data for Language Models
Paper • 2404.07503 • Published • 24 -
Rho-1: Not All Tokens Are What You Need
Paper • 2404.07965 • Published • 80 -
Learn Your Reference Model for Real Good Alignment
Paper • 2404.09656 • Published • 80 -
Dataset Reset Policy Optimization for RLHF
Paper • 2404.08495 • Published • 8 -
Insights into Alignment: Evaluating DPO and its Variants Across Multiple Tasks
Paper • 2404.14723 • Published • 9 -
RLHF Workflow: From Reward Modeling to Online RLHF
Paper • 2405.07863 • Published • 62 -
OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework
Paper • 2405.11143 • Published • 33 -
Mixtures of Experts Unlock Parameter Scaling for Deep RL
Paper • 2402.08609 • Published • 34 -
Scaling Laws for Reward Model Overoptimization in Direct Alignment Algorithms
Paper • 2406.02900 • Published • 10 -
Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs
Paper • 2402.14740 • Published • 6 -
HelpSteer2: Open-source dataset for training top-performing reward models
Paper • 2406.08673 • Published • 12 -
Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback
Paper • 2406.09279 • Published • 1 -
Understanding the performance gap between online and offline alignment algorithms
Paper • 2405.08448 • Published • 11