wop/XXXXXL-chain-of-thought
收藏资源简介:
该数据集名为XXXXXL-chain-of-thought,专注于生成具有思维链(Chain-of-Thought)的AI对话数据,特别强调三种推理模式:直觉(快速、模糊的初始思考)、冷启动(将上下文校准为稳定意识)和热启动(逐步解决问题)。数据集包含单轮和多轮对话示例,但不涉及工具调用。它旨在训练AI模型以更人性化、沉浸式的方式进行推理,同时保持程序严谨性,通过使用思维标签(如think和/think)来展示内部思考过程。数据通过特定提示词合成生成,适用于评估和提升大型语言模型的推理能力。
The dataset named XXXXL-chain-of-thought focuses on generating AI dialogue data with chain-of-thought reasoning, emphasizing three modes: intuition (fast, vague initial thinking), cold start (calibrating all context into a stable consciousness), and hot start (step-by-step problem-solving). It includes examples of single-turn and multi-turn conversations but does not involve tool calling. The dataset aims to train AI models to reason in a more human-like, immersive manner while maintaining procedural rigor, using think tags (e.g., think and /think) to display internal thought processes. Data is synthetically generated via specific prompts and is suitable for evaluating and enhancing the reasoning capabilities of large language models.




