Arena-Hard-Auto Benchmark

What is Arena-Hard-Auto?

Arena-Hard-Auto is an automated benchmark for instruction-tuned LLMs, designed to efficiently approximate human preferences.

Released

2024-04-19

Evaluates

安全, Safety, 大语言模型, LLM, 安全对齐, Safety Alignment, 指令遵循, Instruction Following, 不支持, Unsupported

Openness

open

Importer

OpenCompass Hub

Reported scores

0

Source provenance

• Original evidence https://hub.opencompass.org.cn/dataset-detail/Arena-Hard-Auto

Arena-Hard-Auto paper, code and dataset

• Paper https://arxiv.org/abs/2406.11939
• Code https://github.com/lmarena/arena-hard-auto
• Dataset https://huggingface.co/datasets/lmarena-ai/arena-hard-auto

Which sources cite Arena-Hard-Auto?

• Arena-Hard-Auto

Related 安全, Safety, 大语言模型, LLM, 安全对齐, Safety Alignment, 指令遵循, Instruction Following, 不支持, Unsupported benchmarks

• A-Bench
• A-OKVQA
• AbsPyramid
• ActBench
• ActionAtlas
• Ada-LEval
• AECBench
• AFQMC