🏆 CriticBench Objective Leaderboard 🏆

CriticBench comprehensively evaluates 4 critique dimensions of LLMs on 9 widely-used tasks with multiple response qualities.

description Paper code Code description Project Page description Subjective Leaderboard

📝 Notes

  1. Models labeled with 🌍 are API-Based models, while others are open-sourced.
  2. Some models are not optimized for correction and comparison critique dimensions, like Auto-J-13B and UltraCM-13B. Their scores are not recorded, and the overall scores are the average of other dimensions.
  3. The details about how to compute the overall scores can be found in Section 4.3 in our paper.

This webpage template was recycled from here.