HyperAI
Command Palette
Search for a command to run...
Daring-Anteater 指令微调数据集
Daring-Anteater 是由 NVIDIA 于 2024 年发布的一个用于指令微调的综合数据集,相关论文成果为「HelpSteer2: Open-source dataset for training top-performing reward models」,旨在促进可复现性并训练高性能奖励模型。
该数据集包含约 95,000 个样本,其中大部分由 NVIDIA 专有模型和 Mixtral-8x7B-Instruct-v0.1 合成生成,其余样本来源于 FinQA、wikitablequestions 和 Open-Platypus 的商业友好子集。 该数据集涵盖了广泛的 tasks 和场景,主要用于监督微调(SFT)及后续的偏好调整。 数据集以 CC-BY-4.0 许可开源,支持商业用途。
数据集组成
该数据集包含四个主要字段:
- conversations:用户和助手的对话轮次
- mask:不计算损失的轮次(默认为 "User")
- system:系统提示词
- dataset:数据来源
数据集由以下子集组成:
- synthetic_conv:82,450 个样本
- synthetic_roleplay:2,996 个样本
- synthetic_math:3,000 个样本
- synthetic_precise_instruction_following:1,500 个样本
- synthetic_json_format_following:1,499 个样本
- synthetic_complex_instruction:1,500 个样本
- open_platypus_commercial:6,000 个样本
- FinQA:300 个样本
- wikitablequestions:287 个样本
Citation
@misc{wang2024helpsteer2,
title={HelpSteer2: Open-source dataset for training top-performing reward models},
author={Zhilin Wang and Yi Dong and Olivier Delalleau and Jiaqi Zeng and Gerald Shen and Daniel Egert and Jimmy J. Zhang and Makesh Narsimhan Sreedhar and Oleksii Kuchaiev},
year={2024},
eprint={2406.08673},
archivePrefix={arXiv},
primaryClass={id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg' in_archive='cs' is_general=False description='Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.'}
}
此数据集由社区用户贡献,仅用于教育和信息目的。如有任何内容涉及版权侵权,请通过 [email protected] 联系我们,我们将及时审核并删除。