Dataset · hails
MMLU benchmark without training split
hails/mmlu_no_train
A lightweight copy of the English MMLU text benchmark for rapid model evaluation.
What's inside
Inside are English question-answering text tasks. The volume ranges from 10k to 100k examples. This is a copy of the well-known cais/mmlu set, with the auxiliary_train split removed to speed up subset loading.
What you can do with it
- Evaluate general knowledge and logic of language models (LLMs)
- Accelerate benchmark execution in pipelines without extra data generation
- Tasks
- Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 65.8K
- Likes
- 39
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("hails/mmlu_no_train")