RBench

RBench is a benchmark dataset used to evaluate ranking performance across various tasks and candidate scales, specifically designed to assess the effectiveness of ranking models like LRanker in handling large candidate pools.

Papers using RBench (7)

Goal Alignment in LLM-Based User Simulators for Conversational AI2025 · 14 citesReinforcement World Model Learning for LLM-based Agents2026PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling2025 · 4 citesLRanker: LLM Ranker for Massive Candidates2026CDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling2026Ψ-Bench: Evaluating Persona-Sensitive Influencing in Persuasive Dialogues2026Planner-R1: Reward Shaping Enables Efficient Agentic RL with Smaller LLMs2025