相关数据集
ansulev/opus-4.7-reasoning-cot
Opus 4.7链式思考推理数据集包含2,405个由`claude-opus-4-7`生成的链式思考推理轨迹,涵盖了数学、科学和形式学科等难题。每个样本包括问题→` `块→答案对,其中` `块包含模型的完整思考过程(重述→方法→逐步推导→验证),而答案部分则以独立课程的形式呈现,以粗体结果开头。数据集通过严格的生成和质量控制流程创建,包括模型生成、质量门控和保留策略。数据集
Hugging Face2026-04-24 更新140
ansulev/Hunter-Alpha-SFT-300000x
--- license: apache-2.0 language: - en tags: - conversational size_categories: - 100K<n<1M --- ------------------------------------ # Hunter-Alpha-xhigh-300000x - 308,000 reasoning traces di
Hugging Face2026-03-26 更新50
ansulev/gpt-5.4-step-by-step-reasoning
--- license: mit --- # Dataset Card for GPT-5.4-Reasoning-1500-Ultra-Logic ## Dataset Details ### Dataset Description ### Suggestion: I would use this to fine-tune qwen3.5 35b a3b moe, or 27b varia
Hugging Face2026-04-01 更新40
ansulev/qwen3.6-plus-high-reasoning-500x
--- license: apache-2.0 language: - en tags: - distillation - reasoning size_categories: - n<1K task_categories: - text-generation --- This dataset was prepared for distillation using the **Qwen3.6-p
Hugging Face2026-04-08 更新70
ansulev/GLM-5.1-OpenThoughts3-Distill
GLM-5.1-OpenThoughts3-Distill是一个蒸馏推理数据集,由GLM-5.1模型基于OpenThoughts3-1.2M提示词生成,涵盖科学(物理、化学、生物学等)、代码(编程、算法)和数学(竞赛数学、证明、代数)领域。每个数据行包含用户问题、模型的思维链(thinking)和最终响应(response),并经过质量评估和过滤,保留高质量样本。数据规模包括科学56,974行、代
Hugging Face2026-05-15 更新50



