---
dataset_info:
  features:
  - name: question_id
    dtype: string
  - name: category
    dtype: string
  - name: cluster
    dtype: string
  - name: turns
    list:
    - name: content
      dtype: string
  splits:
  - name: test
    num_bytes: 251691
    num_examples: 500
  download_size: 154022
  dataset_size: 251691
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
license: apache-2.0
task_categories:
- text-generation
language:
- en
tags:
- benchmark
pretty_name: Arena Hard from LMSYS
size_categories:
- n<1K
---

# Arena Hard from LMSYS

This dataset contains the questions used to generate answers using LLMs to evaluate their responses
later on as part of the Arena Hard Benchmark.

The contents in this dataset are originally available in the `question.jsonl` file within
[lm-sys/arena-hard-auto](https://github.com/lm-sys/arena-hard-auto/tree/main/data/arena-hard-v0.1).

## Disclaimer

This dataset is just a re-upload from https://github.com/lm-sys/arena-hard-auto/blob/main/data/arena-hard-v0.1/question.jsonl
for better usability within the Hugging Face Hub, and all the credits and licensing related issues should
go to [LMSYS](https://lmsys.org/) as they are the creators of the dataset and own all the credits.

Anyway, I've filled an issue asking for permission to re-upload this dataset to the Hugging Face Hub, still
under review at https://github.com/lm-sys/arena-hard-auto/issues/19.

In case the authors want me to remove / transfer this dataset I'll do so, so don't rely in this dataset being
here for long until the authors reply to that issue.