AlpacaEval 2.0

AlpacaEval 2.0 is a benchmark used to evaluate the performance of reinforcement learning models by providing a set of tasks that assess their ability to generate high-quality responses based on human feedback.

Papers using AlpacaEval 2.0 (8)

Representation-Aware Advantage Estimation: Your Reward Model Provides More Than A Scalar Output2026Label-Free Reinforcement Learning via Cross-Model Entropy2026AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning2025On the Robustness of Reward Models for Language Model Alignment2025Reward Shaping to Mitigate Reward Hacking in RLHF2025REBEL: Reinforcement Learning via Regressing Relative Rewards2024 · 2 citesSelf-Exploring Language Models: Active Preference Elicitation for Online Alignment2024Just Say What You Want: Only-prompting Self-rewarding Online Preference Optimization2024