opus-research/opus-thinking-10k

Opus Thinking 10k是一个高质量的人工合成数据集，旨在教授大型语言模型（LLMs）链式思考（CoT）推理。该数据集扩展了原始的概念验证数据集，包含超过9,000个示例，涵盖15个推理领域，具有更高的多样性和质量。每个示例遵循严格的格式，模型在回答前会进行“思考”。数据集包括多种推理任务，如编码与调试、科学推理、数学与逻辑、历史与分析、伦理困境、创意写作、常识验证、多步骤任务等。数据集使用MIT许可证，可用于研究和商业用途。

Opus Thinking 10k is a high-quality synthetic dataset designed to teach Large Language Models (LLMs) Chain-of-Thought (CoT) reasoning. This version expands to 9,000+ examples across 15 reasoning domains with improved diversity and quality. Every example follows a strict format where the assistant "thinks" before answering. The dataset includes diverse reasoning tasks such as Coding & Debugging, Scientific Reasoning, Math & Logic, History & Analysis, Ethical Dilemmas, Creative Writing, General Knowledge, Multi-step Tasks, and more. The dataset is licensed under MIT License and is free for research and commercial use.