KZ-Media-Developers/Chronos-Thinking-v1-mini
收藏资源简介:
**Chronos-Thinking-v1-mini** 是一个基础且高密度的数据集,旨在初始化大型语言模型(LLM)中的深度推理过程。该数据集是Chronos Super-AI项目的第一步,不同于自动生成的大规模数据集,v1-mini注重**绝对的知识质量和密度**。它训练模型不仅回答问题,而且像系统架构师一样*思考*,在生成最终答案前使用严格的逻辑。数据集的主要特点是实现了一个严格结构化的`<think>`块,模型通过四个强制认知过滤器解构任务:1. **`[DOMAIN]`**:定义领域;2. **`[ANALYSIS]`**:深度分析任务;3. **`[FORMALISM]`**:数学或逻辑推理;4. **`[LIMITS]`**:解决方案的适用性限制。数据集支持七种语言,格式为JSONL,适用于监督微调(SFT)。
**Chronos-Thinking-v1-mini** is a fundamental, high-density dataset designed to initialize deep reasoning processes in large language models (LLM). This dataset is the first step in the Chronos Super-AI project. Unlike mass datasets generated automatically, v1-mini relies on **absolute quality and density of knowledge**. It trains the model not just to answer questions, but to *think* like a system architect, using strict formal logic before generating the final answer. The main feature of the dataset is the implementation of a rigidly structured `<think>` block. The model is trained to deconstruct any task through 4 mandatory cognitive filters: 1. **`[DOMAIN]`**: Domain definition; 2. **`[ANALYSIS]`**: Deep analysis of the task; 3. **`[FORMALISM]`**: Mathematical or logical reasoning; 4. **`[LIMITS]`**: Awareness of the limits of the applicability of the solution. The dataset is multilingual, supporting seven languages, and is presented in JSONL format, ideal for Supervised Fine-Tuning (SFT).




