llm-traces

For dataset file metadata and data files, please visit the "Files and versions" tab. The current dataset card uses the default template, and the dataset contributor has not provided a more detailed dataset introduction. However, you can download the dataset using the following GIT Clone command or through the ModelScope SDK. A repository containing reasoning traces from LLMs. - Claude reasoning traces were copy-pasted from interface. - Codex is from sessions in app. - All data was extracted without pay, all free to use. - Most Claude reasoning traces were extracted automatically using Selenium via UI. - Claude = Claude Sonnet 4.6, unless specified otherwise. Model responses can have errors. - Some are missing <thinking></thinking> tags, that is fine. I sometimes use Sonnet non-thinking. 这是一个包含大语言模型（LLM）推理轨迹的仓库。 * Claude 的推理轨迹是从界面上复制粘贴的。 * Codex 的数据来自应用内的会话。 * 所有数据均为无偿提取，且可免费使用。 * 大部分 Claude 的推理轨迹是使用 Selenium 自动提取的。 * Claude 4.6 Sonnet