Kukedlc/slopus-v3-train-lt2k

slopus-v3-train-lt2k是一个包含17121个预格式化示例的数据集，专为推理和思维链任务设计。数据使用Qwen3-4B-Thinking-2507模板进行格式化，并经过过滤（确保token数小于2048）和去重处理。数据来源于五个公开子集：roman（7515例，来自Roman1111111/claude-opus-4.6-10000x的reasoning字段）、jirafa（6644例，来自angrygiraffe/claude-opus-4.6-4.7-reasoning-8.7k的think inline字段）、nohurry（2233例，来自nohurry/Opus-4.6-Reasoning-3000x-filtered）、teich（639例，来自TeichAI/Claude-Opus-4.6-Reasoning-887x的thinking字段）和gpt55（90例，来自armand0e/gpt-5.5-chat）。每行数据包含text（完整渲染的聊天模板，包括系统、用户和助手消息，使用<think>...</think>格式）、n_tokens（token计数）、source（来源数据集）和category（源元数据中的类别）。该数据集适用于监督微调（SFT）和推理任务，可通过HuggingFace的datasets库直接加载使用。slopus-v3-train-lt2k is a dataset containing 17,121 pre-formatted examples, designed for reasoning and chain-of-thought tasks. The data is formatted using the Qwen3-4B-Thinking-2507 template, filtered to have fewer than 2048 tokens, and deduplicated. It is sourced from five public subsets: roman (7,515 examples, from Roman1111111/claude-opus-4.6-10000x, reasoning field), jirafa (6,644 examples, from angrygiraffe/claude-opus-4.6-4.7-reasoning-8.7k, think inline field), nohurry (2,233 examples, from nohurry/Opus-4.6-Reasoning-3000x-filtered), teich (639 examples, from TeichAI/Claude-Opus-4.6-Reasoning-887x, thinking field), and gpt55 (90 examples, from armand0e/gpt-5.5-chat). Each row includes text (the fully rendered chat template with system, user, and assistant messages using <think>...</think> format), n_tokens (token count), source (origin dataset), and category (category from source metadata). The dataset is suitable for supervised fine-tuning (SFT) and reasoning tasks, and can be loaded directly via the HuggingFace datasets library.