zail-ai/Qwen_mmlu_dataset

Qwen MMLU基准数据集是一个清洗和去重后的原始MMLU（大规模多任务语言理解）基准版本，旨在确保高质量的、适用于模型的样本，用于微调和评估。该数据集移除了重复条目和不一致的格式，验证了答案键并标准化了主题标签，确保了所有数据分割的一致性，适用于多语言和多任务LLM的微调、推理和特定领域知识的基准测试、学术研究和可再现性研究。

This dataset is a cleaned and deduplicated version of the original MMLU (Massive Multitask Language Understanding) benchmark, curated for high-quality, model-ready samples for fine-tuning and evaluation. It removes duplicate entries and inconsistent formatting, validates answer keys, standardizes subject labels, and ensures a uniform structure across all splits, suitable for fine-tuning multilingual and multitask LLMs, benchmarking reasoning and domain-specific knowledge, and academic research and reproducibility studies.