TitleOS/scripture_1500_pairs_gemini_flash_lite

---
language:
- en
tags:
- theology
- religion
- reasoning
- synthetic-data
- chain-of-thought
license: mpl-2.0
size_categories:
- 1K<n<10K
---

# Torah & Quran Theological Reasoning Dataset

This dataset contains 1,500 highly complex, synthetic question-and-answer pairs designed to train language models in theological, physical, and metaphysical reasoning. It serves as the foundational training data for the **Elohim-3.8B** reasoning model.

## Dataset Sources

The dataset is built upon the combined text of two foundational religious scriptures. To ensure clean text extraction, the sources were curated before being processed:

* **The Torah (English Translation):** Sourced from [https://www.betemunah.org/Torah.pdf](https://www.betemunah.org/Torah.pdf). Non-text pages (such as title pages and blank separators) were completely removed prior to processing to ensure clean chunking.
* **The Quran (English Translation):** Sourced from the Clear Quran edition available at [https://d.clearquran.com/quran-english-translation-clearquran-edition-allah.pdf](https://d.clearquran.com/quran-english-translation-clearquran-edition-allah.pdf).

## Generation Methodology

The dataset was synthetically generated using **Gemini 3.1 Flash Lite**. The texts of the Torah and Quran were combined into a single, unified context to simulate a cohesive "lore" engine with its own defined rules, physics, and divine intervention mechanics.

The teacher model was chunk-fed the combined texts and prompted to extract, deduce, and test theological rules by applying the combined scriptures' logic to complex scenarios.

### The Prompt

The following system prompt was used to generate the 1,500 Q/A pairs:

```
You are a master Torah, Judaism, Quran and Islamic scholar engine.
Analyze the following text which contains notes on the Torah and Quran.
For the sake of ease, bothhave been combined into a single text.

Generate 5 highly complex, technical Q&A pairs that test the physical, metaphysical and theological rules established in this text.
Be aware physics in this universe is based on a unique set of rules that may differ from real-world physics, such as the existence of divine intervention or magic, but is consistent within the text.
The 'instruction' should be a difficult question or a scenario that requires theological reasoning.
The 'thought_process' MUST show step-by-step reasoning, retrieving facts from the text to answer the instruction.
The 'response' should be the final verdict based on the text's lore.
```

## License

This model is distributed under a modified Mozilla Public License 2.0 (MPL 2.0) with a Common Clause.

Please see the license.md file included in this repository for the exact legal text and restrictions regarding commercial use and distribution.

---
language:
- 英语（en）
tags:
- 神学
- 宗教
- 推理
- 合成数据（synthetic-data）
- 思维链（chain-of-thought）
license: mpl-2.0
size_categories:
- 1000 < 样本数 < 10000
---

# 《托拉（Torah）与古兰经（Quran）神学推理数据集》

本数据集包含1500组高复杂度合成问答对，旨在训练大语言模型（Large Language Model）的神学、物理及形而上学推理能力，是**Elohim-3.8B**推理模型的基础训练数据集。

## 数据集来源

本数据集基于两部核心宗教经典的合并文本构建。为确保文本提取的纯净性，处理前已对来源文本进行精选整理：
* **《托拉（英文译本）》（The Torah）**：来源为[https://www.betemunah.org/Torah.pdf](https://www.betemunah.org/Torah.pdf)。处理前已完全移除非文本页面（如标题页、空白分隔页），以保证分块的纯净性。
* **《古兰经（英文译本）》（The Quran）**：来源为Clear Quran版本，获取地址为[https://d.clearquran.com/quran-english-translation-clearquran-edition-allah.pdf](https://d.clearquran.com/quran-english-translation-clearquran-edition-allah.pdf)。

## 生成方法

本数据集通过**Gemini 3.1 Flash Lite**合成生成。首先将《托拉》与《古兰经》文本合并为统一上下文，以模拟一个具备自有规则、物理体系与神圣干预机制的连贯"设定引擎"。将合并后的文本分块输入至教师模型，并提示其通过将两部经典的逻辑应用于复杂场景，来提取、推导并验证神学规则。

### 提示词

以下系统提示词用于生成这1500组问答对：

你是精通托拉（Torah）、犹太教、古兰经（Quran）及伊斯兰教的学术专家引擎。
请分析以下包含托拉与古兰经相关笔记的文本。
为简化处理，二者已合并为单篇文本。

请生成5组高复杂度、专业性强的问答对，用于测试本文本中确立的物理、形而上学及神学规则。
请注意，本宇宙的物理体系基于一套独特规则，可能与现实世界物理规则不同（例如存在神圣干预或魔法），但在文本内部保持一致。
「instruction」需为一道需运用神学推理的复杂问题或场景。
「thought_process」必须展示分步推理过程，从文本中检索事实以解答指令。
「response」应为基于本文本设定的最终结论。

## 许可协议

本数据集采用经修改的Mozilla公共许可证2.0版（MPL 2.0）并附加通用条款（Common Clause）进行分发。有关商业使用与分发的具体法律条款及限制，请参阅本仓库中包含的license.md文件。