HuggingFaceTB/CoT_Reasoning_Bushcraft_Survival

--- dataset_info: features: - name: id dtype: string - name: messages list: - name: content dtype: string - name: role dtype: string splits: - name: train num_bytes: 633962 num_examples: 3000 download_size: 576635 dataset_size: 633962 configs: - config_name: default data_files: - split: train path: data/train-* --- # CoT Reasoning — Bushcraft & Survival This dataset is derived from [`mattwesney/CoT_Reasoning_Bushcraft_Survival`](https://huggingface.co/datasets/mattwesney/CoT_Reasoning_Bushcraft_Survival), reformatted into a conversational prompt-only structure for downstream use (e.g. synthetic data generation, evaluation, or completion by an LLM). ## Origin The source dataset contains 3,000 chain-of-thought Q&A pairs on bushcraft and wilderness survival topics — shelter building, fire starting, water purification, navigation, foraging, tool crafting, and more. Each entry includes a question, a detailed answer, and metadata with difficulty ratings and reasoning traces. ## What changed Only the `id` and `question` columns were kept. The `question` was wrapped into a single-turn `messages` list following the ChatML convention: ```json [{"content": "<question text>", "role": "user"}] ``` The `answer` and `metadata` columns were dropped. See [`create_dataset.py`](./create_dataset.py) for the exact transformation. ## Schema | Column | Type | Description | |---|---|---| | `id` | `string` | Unique identifier from the source dataset | | `messages` | `list[{content, role}]` | Single-turn user message containing the question | ## Example ```json { "id": "qa_20250424_065355_653056", "messages": [ { "content": "What are the crucial considerations when selecting a site for building a natural shelter in a wilderness survival scenario, and what are the three basic types of natural shelters a survivor could construct?", "role": "user" } ] } ``` ## Stats - **Rows:** 3,000 - **Split:** traindataset_info: 特征项: - 名称: id 数据类型: string - 名称: messages 列表: - 名称: content 数据类型: string - 名称: role 数据类型: string 划分集: - 名称: train 字节数: 633962 样本数: 3000 下载大小: 576635 数据集总大小: 633962 configs: - 配置名称: default 数据文件: - 划分集: train 路径: data/train-* # 思维链推理（Chain-of-Thought Reasoning，CoT）—— 丛林求生与野外生存 本数据集源自[`mattwesney/CoT_Reasoning_Bushcraft_Survival`](https://huggingface.co/datasets/mattwesney/CoT_Reasoning_Bushcraft_Survival)，经重格式化后仅保留对话提示词结构，以适配下游任务需求（如合成数据生成、模型评估或大语言模型（Large Language Model，LLM）文本补全）。 ## 数据集来源 源数据集包含3000条针对丛林求生与野外生存主题的思维链问答对，涵盖庇护所搭建、生火、净水、导航、觅食、工具制作等多个领域。每条数据均包含问题、详细解答，以及标注难度等级与思维链推理过程的元数据（metadata）。 ## 修改说明 本次仅保留了`id`与`question`两列，并按照ChatML规范将`question`封装为单轮对话的`messages`列表，格式如下： json [{"content": "<问题文本>", "role": "user"}] 同时移除了`answer`与`metadata`列。具体转换逻辑可参考[`create_dataset.py`](./create_dataset.py)文件。 ## 数据结构 | 列名 | 数据类型 | 说明 | |---|---|---| | `id` | `string` | 源数据集提供的唯一标识符 | | `messages` | `list[{content, role}]` | 包含问题的单轮用户对话消息 | ## 示例 json { "id": "qa_20250424_065355_653056", "messages": [ { "content": "在野外生存场景中选择天然庇护所搭建场地时有哪些关键考量因素，求生者可搭建的三类基础天然庇护所分别是什么？", "role": "user" } ] } ## 统计信息 - **样本条数：** 3000 - **划分集：** 训练集（train）