jade-phoenix-training
收藏资源简介:
Jade Phoenix Training Dataset 是一个专为训练全模态人工智能模型“玉凤”而构建的大规模、多领域数据集。该数据集共包含173,005个唯一文档,旨在为模型提供广泛的知识和语言理解基础。数据来源极其多样化,涵盖了24个不同的领域,包括宗教经典(如圣经、古兰经、薄伽梵歌)、参考指令集(如OpenAssistant、Dolly、Alpaca)、维基百科条目、学术论文(来自CORE和OpenAlex)、科学预印本(arXiv)、网络俚语(Urban Dictionary)、职业技能分类(ESCO)、电子书、医学问答(PubMed QA)、数学问题(GSM8K)、社交媒体内容(如Reddit和4chan)、编程问题(MBPP)、本地数据、GitHub仓库、技术文档(如Rust和StackOverflow)、视觉描述(Wikimedia Commons)、音频样本(Freesound)、人类行为研究、创意艺术、教育理论、心理学主题、政府政策文件、生活指南(WikiHow)以及各类事实与统计数据。这种跨领域的构成使得数据集不仅包含纯文本,还涉及视觉和音频模态的描述,适用于大规模语言模型训练、多领域知识问答、指令遵循、代码生成及多模态理解等多种文本生成任务。
The Jade Phoenix Training Dataset is a large-scale, multi-domain dataset specifically constructed for training the full-modal artificial intelligence model Jade Phoenix. It contains a total of 173,005 unique documents, designed to provide the model with a broad foundation of knowledge and language understanding. The data sources are extremely diverse, covering 24 different domains, including religious classics (such as the Bible, Quran, Bhagavad Gita), reference instruction sets (like OpenAssistant, Dolly, Alpaca), Wikipedia entries, academic papers (from CORE and OpenAlex), scientific preprints (arXiv), internet slang (Urban Dictionary), vocational skill classifications (ESCO), e-books, medical Q&A (PubMed QA), mathematical problems (GSM8K), social media content (e.g., Reddit, 4chan), programming problems (MBPP), local data, GitHub repositories, technical documentation (e.g., Rust, StackOverflow), visual descriptions (Wikimedia Commons), audio samples (Freesound), human behavior research, creative arts, educational theories, psychology topics, government policy documents, life guides (WikiHow), and various facts and statistical data. This cross-domain composition ensures the dataset includes not only pure text but also descriptions related to visual and audio modalities, making it suitable for various text generation tasks such as large-scale language model training, multi-domain knowledge Q&A, instruction following, code generation, and multimodal understanding.
数据集概述
- 名称:Jade Phoenix Training Dataset
- 用途:用于训练 Jade Phoenix 全模态 AI 模型
- 样本数量:287,698 个唯一文档
- 语言:英语(en)
- 任务类别:文本生成(text-generation)
- 标签:llm-training、multi-domain、multimodal、jade-phoenix
数据来源与分布
| 来源类别 | 数量 |
|---|---|
| Wikipedia ZIM(本地 12GB ZIM 文件) | 158,522 |
| 宗教文本(圣经、古兰经、薄伽梵歌) | 37,506 |
| 参考数据集(OpenAssistant、Dolly、Alpaca) | 25,799 |
| 学术来源(CORE API + OpenAlex + PubMed) | 17,463 |
| ArXiv 科学论文 | 6,565 |
| Urban Dictionary 俚语定义 | 11,934 |
| ESCO 技能(职业 + 技能) | 4,594 |
| 书籍(Gutendex、Open Library) | 4,542 |
| 健康问答(PubMed QA) | 5,010 |
| 数学(GSM8K、竞赛数学) | 5,011 |
| 社交媒体(Reddit、4chan、Mastodon、HN) | 3,022 |
| 音频(Freesound API) | 2,492 |
| MBPP 编程问题 | 964 |
| 本地数据(书籍、人物、技能) | 1,688 |
| GitHub 仓库 | 300 |
| 技术内容(Rust、StackOverflow) | 1,041 |
| 视觉内容(Wikimedia Commons) | 250 |
| 人类行为(共情、心理学、角色扮演) | 112 |
| 创意艺术(艺术、音乐、电影、文学) | 220 |
| 教育(教学法、评估) | 157 |
| 心理学主题 | 157 |
| 政府政策文档 | 291 |
| WikiHow 指南 | 198 |
| Wikipedia(多样化主题) | 593 |
| 事实数据(统计、冷知识、世界事实) | 130 |
使用方式
可通过 Hugging Face datasets 库加载:
python from datasets import load_dataset dataset = load_dataset("MC7ever/jade-phoenix-training", split="train")





