ethos-training-data-v1
收藏资源简介:
Ethos-32B Training Data v1是由SustainAI Global(德克萨斯州非营利组织,501(c)(3)待批)创建的公开训练数据集,专为Ethos-32B (ESI)模型设计。该数据集包含8,000至25,000个经过精心筛选的指令调优示例,涵盖可持续经济学、可再生能源、绿色金融、公正转型、伦理人工智能推理以及就业优先对齐等多个领域。所有数据均为合成生成,并在纳入前经过数据委员会审核批准。数据集采用结构化格式,每条记录包含id、messages、domain、source_model、generation_method、committee_approved和review_date等字段,详细信息将在schema.json文件中提供。该数据集的主要发布目的是提高模型训练的透明度和公共问责,允许公众查看Ethos模型的训练内容,但根据CC BY-NC-ND 4.0许可证,禁止商业用途和创建衍生数据集。数据集目前处于开发阶段,完整训练集和模式文件尚未发布,代表性样本将在ethos-transparency仓库中单独提供。
Ethos-32B Training Data v1 is a publicly available training dataset created by SustainAI Global (a Texas nonprofit, 501(c)(3) pending), designed for the Ethos-32B (ESI) model. It contains 8,000 to 25,000 carefully curated instruction-tuning examples across domains such as sustainable economics, renewable energy, green finance, just transition, ethical AI reasoning, and jobs-first alignment. All data is synthetically generated and reviewed and approved by a data committee before inclusion. The dataset is structured with fields including id, messages, domain, source_model, generation_method, committee_approved, and review_date, with detailed information provided in a schema.json file. The primary release aims to enhance transparency and public accountability in model training, allowing the public to see what the Ethos model is trained on, but under the CC BY-NC-ND 4.0 license, commercial use and creation of derivative datasets are prohibited. The dataset is currently in development, with the full training set and schema files not yet released; representative samples will be provided separately in the ethos-transparency repository.
数据集概述:Ethos-32B Training Data v1
- 发布机构:SustainAI Global(美国德克萨斯州非营利组织,501(c)(3)状态待批)
- 数据集名称:Ethos-32B Training Data v1
- 许可证:CC BY-NC-ND 4.0(仅用于透明度和公共问责,禁止商业用途和衍生作品)
- 语言:英语
- 标签:伦理AI、可持续经济学、绿色金融、指令微调、合成数据
- 数据规模:10,000 < n < 100,000(目标规模为8,000至25,000个精选样本)
内容描述
该数据集是为训练 Ethos-32B(ESI) 模型而准备的公共指令微调语料库。样本为合成生成,并需经过数据委员会审核后方可纳入。覆盖领域包括:
- 可持续经济学
- 可再生能源
- 绿色金融
- 公正转型
- 伦理AI推理
- “就业优先”(Jobs First)对齐
文件结构
| 文件名 | 说明 |
|---|---|
train.jsonl |
委员会批准的完整训练集(尚未发布) |
schema.json |
字段定义和来源规范(即将发布) |
数据预览
约50至100个代表性样本将单独发布在 ethos-transparency/data/sample,但不替代完整数据集。
数据溯源
每条记录将包含以下字段:id、messages、domain、source_model、generation_method、committee_approved、review_date。详细定义待 schema.json 发布。
当前状态
数据集目前处于 进行中 状态,完整训练数据尚未公开。
联系方式
- 邮箱:ai@sustainai.global
- 官网:https://sustainai.global/
- 透明度文档:https://github.com/SustainAI-Global/ethos-transparency




