Arena-Hard v-0.1

'Arena-Hard v-0.1' is a benchmark dataset used to evaluate the performance of Direct Preference Optimization (DPO) methods in optimizing human preferences for large language models.

Papers using Arena-Hard v-0.1 (2)

Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment2026Pre-DPO: Improving Data Utilization in Direct Preference Optimization Using a Guiding Reference Model2025