MT-Bench

The 'MT-Bench' dataset is a benchmark used to evaluate instruction-following capabilities of large language models through preference data.

Papers using MT-Bench (12)

From Demonstrations to Rewards: Alignment Without Explicit Human Preferences2025 · 4 citesDirect Advantage Regression: Aligning LLMs with Online AI Reward2025 · 1 citesRepresentation-Aware Advantage Estimation: Your Reward Model Provides More Than A Scalar Output2026Reward Model Routing in Alignment2025TGDPO: Harnessing Token-Level Reward Guidance for Enhancing Direct Preference Optimization2025AutoRule: Reasoning Chain-of-thought Extracted Rule-based Rewards Improve Preference Learning2025Pretrain Value, Not Reward: Decoupled Value Policy Optimization2025SimPO: Simple Preference Optimization with a Reference-Free Reward2024 · 17 citesRLHF Workflow: From Reward Modeling to Online RLHF2024 · 3 citesREBEL: Reinforcement Learning via Regressing Relative Rewards2024 · 2 citesAre You Sure? Rank Them Again: Repeated Ranking For Better Preference Datasets2024Self-Exploring Language Models: Active Preference Elicitation for Online Alignment2024