license: apache-2.0
task_categories:
- text-generation
language: - fr
pretty_name: IFEval-FR
IFEval-FR
French split of the M-IFEval instruction-following benchmark, in the same format as google/IFEval. This version was reviewed and cleaned with Claude Opus 4.8 (fixed prompt/verifier mismatches, off-by-one constraints, an untranslated prompt, and grammar/formatting).
from datasets import load_dataset
ds = load_dataset("kurakurai/IFEval-FR")
Fields
key: unique idprompt: the instruction (French)instruction_id_list: verifiable instructions (fr:prefixed)kwargs: arguments for each instruction
235 prompts, single train split. Apache-2.0.
Citation
@article{Dussolle2025MIFEval,
title={M-IFEval: Multilingual Instruction-Following Evaluation},
author={Dussolle, Antoine and Cardena D\'iaz, Andrea and Sato, Shota and Devine, Peter},
year={2025},
eprint={2502.04688},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2502.04688}
}