luojunyu/SemiEvol

Dataset Card for Dataset Name

The SemiEvol dataset is part of the broader work on semi-supervised fine-tuning for Large Language Models (LLMs). The dataset includes labeled and unlabeled data splits designed to enhance the reasoning capabilities of LLMs through a bi-level knowledge propagation and selection framework, as proposed in the paper SemiEvol: Semi-supervised Fine-tuning for LLM Adaptation.

Dataset Details

Dataset Sources [optional]

• Repository:
  • https://huggingface.co/datasets/cais/mmlu
  • https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro
  • https://huggingface.co/datasets/cais/mmlu;
  • https://huggingface.co/AdaptLLM;
• Paper [optional]: https://huggingface.co/papers/2410.14745

Uses

Direct Use

Please refer to https://github.com/luo-junyu/SemiEvol for evaluation instruction.

Citation [optional]

BibTeX:

@misc{luo2024semievol,
    title={SemiEvol: Semi-supervised Fine-tuning for LLM Adaptation}, 
    author={Junyu Luo and Xiao Luo and Xiusi Chen and Zhiping Xiao and Wei Ju and Ming Zhang},
    year={2024},
    eprint={2410.14745},
    archivePrefix={arXiv},
    primaryClass={cs.CL},
    url={https://arxiv.org/abs/2410.14745}, 
}

APA:

Luo, J., Luo, X., Chen, X., Xiao, Z., Ju, W., & Zhang, M. (2024). SemiEvol: Semi-supervised Fine-tuning for LLM Adaptation. ArXiv. https://arxiv.org/abs/2410.14745