Skip to content

Dataset · HuggingFaceH4

MT-bench prompt dataset for chatbot evaluation

HuggingFaceH4/mt_bench_prompts

A benchmark set of English prompts from LMSYS for testing conversational models.

What's inside

Contains English text prompts created by LMSYS to evaluate chat model quality within the MT-bench benchmark. The dataset size is small—less than 1,000 records. In this version, the turns field has been renamed to prompts, and empty references have been added for internal tools.

What you can do with it

  • Test and compare instruction-tuned LLMs in multi-turn dialogues
  • Automate model response evaluation using LLM-as-a-Judge
Tasks
Question answering
Languages
English
Size
up to 1K rows
Downloads this month
19.8K
Likes
27

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("HuggingFaceH4/mt_bench_prompts")