index.md · 可读全文

这里保留原文的中文、公式标记、代码和链接。可直接交给 Agent 读取，也可选取下方原文复制。此入口返回 HTML；原始 .md / .py 文件仍保留，完整候选包同时包含 UTF-8 原件。

# 从同一句话走进自己的问题

主线从“我昨天给了他一本书”出发，用九个教学单位贯通传统NLP、共享表示、RNN/LSTM、完整生成、逐次读取和Transformer。然后明确改变训练条件与任务：移去中文学习普通文本，将指令与中文放入decoder-only的同一因果序列；补入同一事件的图像，并将任务改为带颜色的英文描述，才有a→blue→book。执行部分继续这个生成时刻，再显式切到多query的prefill/训练讨论IO。

这些完整主题有独立定义、机制图、数学与代码，可以从原输入开始独立理解。旧六片段她/her、三行BLOCK、合成反转和历史结果保留原身份；新九单位他/him不覆盖旧checkpoint，也不改变Notebook ID。资源章节完整保留；默认课堂六分钟建立缓存与搬运直觉，可选T60–T65另用十一分钟展开账本。

|从哪里继续|完整正文|原Notebook|
|---|---|---|
|同一句话怎样成为分析对象与模型输入|[主题](https://codingai-lec05.pages.dev/course/knowledge/text-units-and-representations.html)|[CLS01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-01-b1c8e2fd764441e8820b808fc40ec3cb?utm_content=b1c8e2fd764441e8820b808fc40ec3cb&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-01-ec403d28cc2c400ba661cd270588669f?utm_content=ec403d28cc2c400ba661cd270588669f&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|把预测、训练与评价摊在同一张样本表上|[主题](https://codingai-lec05.pages.dev/course/knowledge/objectives-and-evaluation.html)|[CLS01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-01-b1c8e2fd764441e8820b808fc40ec3cb?utm_content=b1c8e2fd764441e8820b808fc40ec3cb&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-01-ec403d28cc2c400ba661cd270588669f?utm_content=ec403d28cc2c400ba661cd270588669f&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|状态在流动，梯度可以停在边界|[主题](https://codingai-lec05.pages.dev/course/knowledge/recurrent-state-and-credit.html)|[CLS01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-01-b1c8e2fd764441e8820b808fc40ec3cb?utm_content=b1c8e2fd764441e8820b808fc40ec3cb&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB01](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-01-ec403d28cc2c400ba661cd270588669f?utm_content=ec403d28cc2c400ba661cd270588669f&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|2014：先拥有一个可工作的翻译器，再决定改哪里|[主题](https://codingai-lec05.pages.dev/course/knowledge/translation-choices-2014.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-02-bba1d024d8e64a589c290b09381de692?utm_content=bba1d024d8e64a589c290b09381de692&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|读取之后，系统怎样决定写 him|[主题](https://codingai-lec05.pages.dev/course/knowledge/attention-to-prediction.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-02-bba1d024d8e64a589c290b09381de692?utm_content=bba1d024d8e64a589c290b09381de692&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|让整套 Transformer 共同完成一次预测|[主题](https://codingai-lec05.pages.dev/course/knowledge/transformer-information-flow.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|同一段文本，三种预训练问题|[主题](https://codingai-lec05.pages.dev/course/knowledge/model-objectives.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|在已经认识的 block 上读懂现代部件|[主题](https://codingai-lec05.pages.dev/course/knowledge/modern-transformer-parts.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|从2017计算图到2020组合，再看持续状态系统|[主题](https://codingai-lec05.pages.dev/course/knowledge/architecture-choices-and-state.html)|[CLS02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-02-Attention-Transformer-2d44adcb6f3d409aa7dfb25c21a748a3?utm_content=2d44adcb6f3d409aa7dfb25c21a748a3&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-02-bba1d024d8e64a589c290b09381de692?utm_content=bba1d024d8e64a589c290b09381de692&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|路径 A：分数进步了，为什么这一句仍然错|[主题](https://codingai-lec05.pages.dev/course/knowledge/teacher-forcing-and-generation.html)|[CLS03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-03-719b25c7c5b64687998ba9c19d2f6776?utm_content=719b25c7c5b64687998ba9c19d2f6776&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-02-bba1d024d8e64a589c290b09381de692?utm_content=bba1d024d8e64a589c290b09381de692&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|路径 B：换图后分布变了，模型是否看懂了|[主题](https://codingai-lec05.pages.dev/course/knowledge/visual-condition-use.html)|[CLS03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-03-719b25c7c5b64687998ba9c19d2f6776?utm_content=719b25c7c5b64687998ba9c19d2f6776&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|路径 C：材料放得下，不代表采用了当前事实|[主题](https://codingai-lec05.pages.dev/course/knowledge/cache-and-effective-context.html)|[CLS03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-03-719b25c7c5b64687998ba9c19d2f6776?utm_content=719b25c7c5b64687998ba9c19d2f6776&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|
|给长前缀列一张有单位的资源账本|[主题](https://codingai-lec05.pages.dev/course/knowledge/resource-ledger.html)|[CLS03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/CLS-03-719b25c7c5b64687998ba9c19d2f6776?utm_content=719b25c7c5b64687998ba9c19d2f6776&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook) → [LAB03](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/Fall-26-Lecture-05-841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/LAB-03-Transformer-9c5b3d45acdb438a8e7df1dc6cb97e40?utm_content=9c5b3d45acdb438a8e7df1dc6cb97e40&utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_term=get_notebook)|

## 和Agent共读

读一段完整正文，提出自己真正的疑问，把主题URL交给Agent；需要复算时再提供样本、配置或输出，然后回同一Notebook验证。无需背术语、复制固定提示词或先运行所有单元。保存结果、只读重评、训练和未来研究设计有各自身份；私人学生输出不会自动公开。