{"schema_version":"1.0","language":"zh-CN","generated_from":"NEWVAR 从零开始学AI structured Markdown","write_access":false,"full_text_access":false,"content_boundary":"This endpoint exposes metadata, objectives, prerequisites, labs and source summaries. It does not expose chapter body text or answer text.","chapter":{"slug":"attention-transformer","title":"注意力与Transformer","english_title":"Attention and transformers","chapter_number":55,"volume":3,"volume_title":"第三卷：主要AI分支","month":14,"unit_title":"自然语言处理","review_status":"draft","updated_at":"2026-08-06","estimated_hours":8,"focus":"从查询、键、值理解内容寻址，掌握多头注意力、位置和因果遮罩。","objectives":["从查询、键、值理解内容寻址，掌握多头注意力、位置和因果遮罩。","解释核心数学关系并完成最小实现","运行单变量实验并记录失败样例","完成工程线或研究线至少一项迁移任务"],"prerequisites":["embeddings-rnn-lstm"],"exercise_count":3,"tracks":["engineering","research"],"human_url":"/learn/attention-transformer"},"labs":[{"id":"lab-attention-transformer","chapter_slug":"attention-transformer","title":"注意力与Transformer：单变量实验","kind":"attention_temperature","data_status":"教学模拟","variable":{"label":"注意力温度","unit":"","min":0.1,"max":3,"step":0.1,"default":1},"objective":"只改变“注意力温度”，观察结果、代价和风险如何一起变化，理解从查询、键、值理解内容寻址，掌握多头注意力、位置和因果遮罩。","static_fallback":"静态替代：把注意力温度分别设为0.1、1和3，手工比较三组教学模拟输出。","keyboard":"聚焦滑块后使用方向键微调，Page Up/Page Down大步调整，Home/End到达边界。","measurement_note":"页面数值由公开公式生成，只用于教学，不代表真实模型性能或现实世界因果效果。","source_ids":["S029","S030","S031","S055"]}],"project":{"id":"project-14","month":14,"title":"做一个带检索证据的校园FAQ原型","brief":"只使用公开校规或虚构资料，完成分词、检索、回答、引用、拒答和错误分析。","deliverables":["问题与边界说明","可运行最小实现","实验记录与失败分析","来源与许可清单","风险、隐私与人工闸门","复现README"]},"sources":[{"id":"S029","author":"Stanford University","title":"CS224N: Natural Language Processing with Deep Learning","source_level":"大学课程一手资料","published_at":"2026","accessed_at":"2026-08-06","doi_or_url":"https://web.stanford.edu/class/cs224n/"},{"id":"S030","author":"Ashish Vaswani et al.","title":"Attention Is All You Need","source_level":"同行评审论文预印本","published_at":"2017","accessed_at":"2026-08-06","doi_or_url":"https://arxiv.org/abs/1706.03762"},{"id":"S031","author":"Jacob Devlin et al.","title":"BERT: Pre-training of Deep Bidirectional Transformers","source_level":"同行评审论文预印本","published_at":"2018","accessed_at":"2026-08-06","doi_or_url":"https://arxiv.org/abs/1810.04805"},{"id":"S055","author":"Patrick Lewis et al.","title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","source_level":"同行评审论文预印本","published_at":"2020","accessed_at":"2026-08-06","doi_or_url":"https://arxiv.org/abs/2005.11401"}]}