| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
fix(i18n,ch2): 13 个译本同步图 2-10 缓存失效修正 (#1033) | 12 天前 | |
refactor(i18n): sync the Chapter 6 "Interaction" restructure to all 12 translations (#859) * fix(figures): close Chinese numbering gaps and rebuild fig8-1 The Chinese source had reader-visible figure numbering gaps left behind by earlier content removals, and the caption number no longer matched the filename in two chapters. - ch5: restore the 图5-4 embed dropped by d39b1d74 while its prose reference survived, closing the 5-3 -> 5-5 gap. - ch7: retire the two figures the chapter dropped to descriptive names and shift fig7-18/19/20 down so caption number == filename again. - ch10: renumber 10-3..10-13 to 10-2..10-11 across all 13 editions, closing the gaps at 10-2 and 10-10; retire the two removed figures to descriptive names. Drop the dangling "Figure 10-10" reference in the English edition. - Rebuild fig8-1 for the 11 editions still shipping the pre-04c0ed70 artwork. The Chinese source now has gapless, filename-aligned numbering in every chapter with no dangling cross-references. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): rebuild the stale chapter 8 artwork in 11 editions Commit 04c0ed70 replaced all five Chinese chapter-8 figures when the chapter was reorganized around continual evolution. Only book-ko, added afterwards, picked them up; every other edition still shipped the pre-reorganization artwork, so the captions described the new chapter while the pictures showed the old one. In book-en, book-ar and book-zhtw fig8-3 was additionally still untranslated Simplified Chinese, and book-es/ja/tr carried a further two-figure offset whose fig8-3/fig8-4 showed chapter-4 material. Rebuilds fig8-1..fig8-5 for all 11 affected editions from the current Chinese geometry, translating text nodes only so the drawings stay identical, and drops the orphaned fig8-6/fig8-7 left in book-es, book-ja and book-tr. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): rebuild wrong and stale chapter 3 artwork in 8 editions Eight editions still carried chapter-3 figures from an earlier revision, two of which showed the wrong drawing entirely: - fig3-2 ("four memory strategies") rendered the RAG query flow, i.e. the artwork belonging to fig3-5. - fig3-4 ("multi-type memory architecture") rendered the HNSW index structure from fig3-7. - fig3-1 showed the retired RAG-fundamentals map instead of the chapter knowledge map (fixed for book-en in #816, never carried across). - fig3-11 showed the retired GraphRAG walkthrough built on the x86 SSE/AVX instruction set -- a CPU diagram in a knowledge-graph section. In book-hu it instead showed fig3-14's contextual retrieval. - fig3-12 was an earlier revision of the agentic vs. non-agentic comparison. All five are regenerated for ar, es, hu, id, ja, ru, ta and tr from the current book-en geometry, translating text nodes only, so every edition now matches the Chinese drawing exactly. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): align chapter 9 caption numbering and rebuild fig4-9 Every edition captioned fig9-7 as "9-6", duplicating the number already used by fig9-6, so all later captions ran one behind their file and the chapter ended with fig9-12 labelled 9-11. Captions are now renumbered to match their filenames, and the fig9-11/fig9-12 embeds (VLA, Sim2Real) are dropped to match the Chinese chapter, which removed them in #730. The files stay on disk; the course slides link fig9-11 directly. Also rebuilds fig4-9 from the Chinese for the six editions that shipped it untranslated or, in book-es and book-tr, showing a duplicate of fig4-8, and switches book-hu to the shared fig2-7.png attention heat map that the other twelve editions use. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): renumber chapter 9 captions to match their files Every edition captioned fig9-7 as "9-6", duplicating the number already used by fig9-6, so each later caption ran one behind its file and the chapter ended with fig9-12 labelled 9-11. Captions now match their filenames in all twelve editions; the figure set itself is unchanged, so the prose cross-references still resolve. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): renumber the remaining Russian chapter 10 caption The chapter 10 renumber matched captions on the "Рис." prefix and missed the one caption written as "Рисунок 10-11", leaving it pointing at fig10-9. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(i18n): sync chapter 9 prose to the Chinese and fix Hungarian ordering Chapter 9's numbering gaps at 9-5/9-6 were a prose problem, not a figure problem: eight editions had condensed the "cognitive timing" section into a single summary paragraph, so the fast/slow-thinking and Step-Audio R1 figures had nowhere to attach. Their SVGs were already present and correctly translated -- just never referenced. - Restore the three #### subsections the Chinese has (fast thinking answers / fast thinking interacts / end-to-end unification) in ar, es, hu, id, ru, ta, tr and vi, translated to match each edition's existing terminology, and re-attach fig9-5 and fig9-6. - Drop the fig9-11/fig9-12 embeds and their in-text references in all twelve editions: #730 deliberately removed both from the Chinese chapter, and the Chinese is authoritative. The VLA and Sim2Real prose stays. - book-hu: move the fig7-7 experiment box back ahead of the pre-training section and the fig10-9 MetaGPT section after fig10-8, matching the Chinese reading order. Chapter 9 is now exactly ten figures in all thirteen editions, and every chapter of every edition is gapless, caption-aligned and in order. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(i18n): sync the chapter 7 reward section to the Chinese (en, zhtw, ja, es, ru, tr) Commit 05ab7f3e condensed the Chinese reward-design section into four subsections; the translations still carried the superseded version, roughly ten times longer and organised differently, along with fig7-16 and fig7-17 which the Chinese no longer has. Replaces the three old reward subsections with a translation of the current Chinese four-subsection structure and adds the [^ch7-23] and [^ch7-29] footnotes, which existed only in the Chinese. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(i18n): sync the chapter 7 reward section across all 12 editions Commit 05ab7f3e condensed the Chinese reward-design section from three loosely organised subsections into four sharper ones and dropped two figures. The translations never followed, so every edition still shipped the superseded version -- roughly ten times longer than the Chinese and structured differently -- along with fig7-16 and fig7-17, which no longer exist upstream. - Replace the old reward subsections in ar, en, es, hu, id, ja, ko, ru, ta, tr, vi and zhtw with a translation of the current Chinese four-subsection structure, matching each edition's established terminology. - Add the [^ch7-23] and [^ch7-29] footnotes, previously Chinese-only. - Retire fig7-16/fig7-17 to descriptive names and shift fig7-18/19/20 down, so chapter 7 is 18 figures with caption number == filename in every edition, exactly as the Chinese source now is. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(figures): translate the remaining Chinese text in chapter 4 figures fig4-7 and fig4-8 shipped with untranslated Chinese labels in book-en, book-ar, book-ta and book-vi; the Arabic copies were additionally part-machine-translated into garbled mixtures. Both are rebuilt from the Chinese geometry with proper translations, keeping tool identifiers and protocol role names as-is. Also translates the one English sentence left in book-ta's fig9-4. The Chinese text remaining in fig2-6 and fig10-5 is deliberate: it is the example sentence being analysed and the glossary data the figure illustrates. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(i18n): align experiment/figure/table numbering across all 12 translations Bring the numbering skeleton of every translated edition into line with the Chinese edition, so that later structural work can be applied mechanically. The Chinese edition is the source of truth throughout. Experiments - ch2: add the missing 实验 2-7 (writing Skill from personal samples) to all 12 editions and shift the following experiments 2-7..2-9 → 2-8..2-10; rename the accompanying footnote ch2-7 → ch2-8 to match. - ch4: add the missing multimodal-extraction experiment as 4-2 to 10 editions and shift 4-2..4-6 → 4-3..4-7. book-en had this experiment mislabelled as "Experiment 3-7" (colliding with a real 3-7 in chapter 3) and placed before 4-1; it is relabelled 4-2 and moved after the 4-1 box. Chapter summaries that enumerate the experiments are updated (six → seven, ranges corrected). - ch5: book-hu had the production-log experiment mislabelled 5-5 (duplicating the real 5-5) and was missing 5-7 entirely; the box is renumbered 5-8 and a translation of 实验 5-7 (adaptive log parser) is added. - ch7: add the missing 7-12 (V-IRL-VL) and 7-13 (SimpleVLA-RL) boxes to all 12 editions, together with the [^ch7-24] footnote they cite. Also fix the stale in-text numbers: ReTool was cited as 7-15 (it is 7-14) and RLVP as 7-14 (it is 7-16). Figures - ch9: 8 editions compressed the fast/slow-thinking section into a single paragraph and therefore lost 图9-5 and 图9-6. The three-solution block is translated in full so all editions carry the same figure inventory. Tables - ch6: remove the translation-only Pass@k/Pass^k table (no counterpart in the Chinese) and renumber 6-4..6-6 → 6-3..6-5. In book-en this table also duplicated the number of the memory-system table. - ch10: remove the translation-only shared/non-shared selection-criteria table and renumber 10-2..10-4 → 10-1..10-3. - ch9: add the 表9-1 caption to the MiniCPM-o experiment in 11 editions. - book-id ch3: translate the untranslated "Table 3-3" caption. - book-ar ch7: fix the transposed table number 1-7 → 7-1. Other consistency fixes - book-hu: 542 headings and labels across all chapters had their bold markers replaced by straight double quotes, which hid whole experiment boxes from the numbering. Restored (code blocks left untouched; the edition uses „…" for real quotations). - book-hu ch10: move the 10-4 experiment box back into the manager-pattern section so 图10-8 precedes 图10-9. - book-ar ch2: drop an Arabic-only passage (and its [^ch2-5] footnote) that has no counterpart in the Chinese. - book-ru ch2: add the missing [^lost-in-the-middle] footnote. - book-vi ch1: Thử nghiệm 1.2 → 1-2. After this change the only remaining gap is book-ar/chapter4, which is covered by the separate chapter-4 retranslation PR. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * refactor(i18n): sync the Chapter 6 "Interaction" restructure to all 12 translations Mirror the Chinese restructure (#856) in every translated edition, so that all 13 editions share one chapter layout, numbering scheme and file naming. Structure - New Chapter 6 "Interaction: Expanding the Observation and Action Spaces", assembled from a newly written framing section, the event-driven async block lifted out of Chapter 4, and the whole of the old Chapter 9 (voice, Computer Use, robotics), followed by a merged summary and 12 merged exercises. - Old chapters 6/7/8 shift to 7/8/9; the old chapter 9 file is consumed by the new chapter 6. Chapter 4 keeps the three Agent-invoked tool categories and hands the event-triggered and user-communication tools to Chapter 6. Mechanical parts, applied per edition with language-specific label patterns - Cross-chapter references rotate 6→7→8→9→6 (including Arabic ordinal words and Traditional Chinese numerals). - Figures, experiments, tables and footnote IDs renumbered with one global map: fig4-2..4-6 → fig6-1..6-5, fig9-N → fig6-(N+5), fig4-7..4-9 → fig4-2..4-4, and 6→7, 7→8, 8→9 for the shifted chapters; 56 SVGs renamed per edition. - Companion code links ( ../chapterN/) rotate with the code directories. New prose written for every edition - Chapter title, opening, the "Two Axes: Modality and Timing" section with both tables, the async section heading and its lead-in, the chapter summary, and the twelfth exercise plus its reference answer. - Chapter 4's roadmap paragraph, tool-classification paragraph and summary are rewritten to hand off the two event-driven tool categories. - introduction and afterword: the part structure becomes 2–6 / 7–9 / 10, the chapter-6 bullet is rewritten and moved into place, the old multimodal bullet is retired, and the reading paths and prerequisite chapter numbers follow. - reference-answers: Chapter 4 drops to four answers, the three that moved plus the eight from the old chapter 9 and one new answer form the chapter 6 section, which is repositioned accordingly. Also fixes seven stale companion links in the Chinese edition that #856 left pointing at the pre-rotation directories. Verified: for all 12 editions × 10 chapters, the experiment inventory, figure inventory and order, and thought-question count match the Chinese exactly; no dangling footnotes, no missing image files, no unbalanced code fences. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 25 天前 | |
docs(book): reorganize chapters for version 2.0 (#909) | 23 天前 | |
docs(i18n,ch1): 同步简化 Agent 结构的设计原则 (#1040) | 9 天前 | |
docs(ch1,ch10): 补充「自主 Agent 生成工作流」这一编排形态 (#1038) 第一章的「两种模式的选择与混合」只讲了工作流节点与自主 Agent 节点在同一 系统中并存这一种混合方式，缺了实践中越来越常见的另一种：先由自主 Agent 把工作流写出来，再由工作流去执行——拓扑由模型现场决定，执行阶段则退回工 作流的确定性。本次在 n8n 一段之后补一段说明，并前向引用第十章。 第十章的管理者模式此前只有顺序协调和并行协调两种形态，两者都要求管理者 Agent 始终待在循环里：每派发一个子任务都要模型做一次决策，上下文随调用 次数增长。本次在实验 10-4 之后补入第三种形态——管理者先把 Agent 工作流 写成一段代码，再交给确定性的运行时执行，并以 Claude Code 内置的 Workflow 工具为例，给出 agent()/parallel()/pipeline() 原语和一段 pipeline 编排骨架 （七组事实各自先调研、再逐条独立核实，最后统一汇总）。 同一改动同步到 ar/en/es/he/hu/id/ja/ko/ru/ta/tr/vi/zh-TW 十三个译本，每个 译本两处：第一章 n8n 图之后一段，第十章图 10-8 实验框之后的整节。术语沿用 各版既有惯例（en 用 sub-agent、es 用 subagente、he 用 תת‑סוכן、ko 用 하위 에이전트、ru 用 дочерний агент 等）；代码注释按各版惯例本地化，ar 与 he 两 个 RTL 版本沿用英文注释。 校验：14 个版本代码围栏成对，scripts/check_i18n_consistency.py 通过。 Claude-Session: https://claude.ai/code/session_0136KBQ5qSzmdU28eybszQvY Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 9 天前 | |
docs(ch2): 修正 KV Cache 例子中查询向量归属与 K、V 组数的错位 (#1056) 第二章「KV Cache 的原理与约束」用 [A, B, C, D] 生成 E 的例子讲解注意力， 其中三处与自回归解码的实际过程不符（issue #1042）： 1. 「E 的查询向量与所有已有 token 的键向量做点积」——生成 E 这一步 E 还 没被采样出来，位置 5 尚不存在，参与计算的是最后一个已知 token D 的 Query，得到的是 D 位置的输出表示，模型用它预测出 E。 2. 「生成 E 时要计算 5 组 K、V」——此时前缀只有 A~D，是 4 组；6 组要到 生成第 6 个 token（前缀含 E）时才出现。整体仍是 O(N²)，但逐项数值错位。 3. 「使用 KV Cache 时……生成 E 时只需要计算 E 自身的 K、V」——同理，这一 步新增进缓存的是 D 的 K、V；E 自身的 K、V 要等 E 被采样并送回模型后 才计算。 改写这三段，明确「每一步只为最新进入上下文的那个 token 计算 K、V」，并补 一句说明 E 自己的 Q、K、V 何时才产生。图 2-10 不涉及该例子，无需改动； 实验 2-2 的热力图描述是对完整序列而言，本身正确，未改。 15 个语种同步。 Fixes #1042 Claude-Session: https://claude.ai/code/session_018iSm7JBWoy87hxSpUkJ49T Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 6 天前 | |
docs(i18n,ch3): 第 3 章 13 个译本全部与中文原稿对齐 (#1011) * docs(i18n,ch3): 第 3 章 13 个译本全部与中文原稿对齐 沿用第 1 章那套两道筛查（逐节段数 + 单元字符量相对基准的比值）。 段数层面 - §「超越扁平文本：知识的组织与检索」：10 个译本多出一段中文没有的 「传统 RAG 的扁平化处理」；en/he/id 则相反，缺开篇那段（RAG 基础技术 解决的是「给定文本块怎么找」，更根本的问题是「文本块本身怎么组织」， 以及本节最后要把这些方法反过来用到用户记忆上） - §「混合检索」：en/he/id 把「三个阶段」引子与「第一阶段并行检索」并成 一段，拆回两段 - §「记忆能力的评估」：ar 把第一层与第二层并在一段 - §「用户记忆的认知科学基础」：ta 缺工作记忆／长期记忆的划分一段 - §「RAG 技巧：上下文感知检索」：es 多出一段中文没有的小结 内容层面（段数相同但被压缩或仍是旧稿） - §「User as Code」开篇：13 个译本都是更长的旧稿（把文本记忆的三步缺陷 与「活的软件工程项目」比喻都写进正文），中文已收短为一句判断，按原稿 收短 - Simple Notes（ar/hu/ko/ru/ta）、Enhanced Notes（hu）：丢掉 O(1) 的解释 或「同一份工作被割裂成三条独立事实」的例子 - Mem0 v2 提取-对比-决策（ru/tr/vi）与 v3 仅追加写入（ar/es/hu/id/ru/ta/vi）： 丢掉 Mem0-g 图记忆变体、LoCoMo/LongMemEval 的具体数字、OSS 已移除外部 图存储因此 Mem0-g 应视为历史设计等 - 神经重排序（ar/hu/ru/ta）：丢掉「它不是为补救 RRF 丢分而存在」「重排序 不替代融合」两处要点 - BM25 参数说明（ar）、RAPTOR 聚类与递归抽象（ar）、何时需要结构化索引 （ko/ru）、失效内容下线与租户隔离（ko/ru）、多模态嵌入存进上下文（ru）、 六个主题的导览（ru） 校验：13 个译本与中文逐节单元数一致；字符量比值除两处 LaTeX 密集单元 （公式占比高，比值天然偏低）外无异常；脚注无悬空定义与未定义引用； check_i18n_consistency.py 与 pytest tests/ 通过；译本无 CJK 混入， zhtw 无简体字。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe * fix(i18n,ch3): 修正 ta 版混入的中文字「筛筛」 双编码器/交叉编码器的类比里，「大规模初筛」的「初筛」在泰米尔语版中 残留成了「筛筛」两个汉字，现改为「முதற்கட்டத் தேர்வு」。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 14 天前 | |
docs(i18n,ch4): 第 4 章 13 个译本全部与中文原稿对齐 (#1012) 沿用两道筛查（逐节段数 + 单元字符量相对基准的比值）。 实验方框归位 - en/he 把实验 4-2（感知工具 MCP 服务器）错放进「工具化多模态分析」 一节，并且整个丢失了实验 4-3（多模态信息提取：三种技术范式的对比 分析）。现把 4-2 移回「感知工具」一节末尾，并按中文原稿补上 4-3 段数层面 - §「Skills：把工具发现变成按需查阅」：12 个译本多出一段与下一段重复 的「渐进式披露」小结 - §「感知工具」：10 个译本把开篇一句拆成两段 - §「提取为文本」：10 个译本把 2 段压成 1 段，丢掉了两阶段流程的说明， 以及「一页 PDF 截图上千 token、文字只有几百 token」这个具体对比 - §「工具化多模态分析」：10 个译本把 3 段压成 2 段，丢掉了工具签名 （analyze_image / analyze_pdf / analyze_audio 接受文件与自然语言问题） 和「内部多模态模型不必有很强 Agent 能力」的选型空间 - §「能力的表达形式」：tr 缺「第九章将讨论持续进化中如何做同样选择」 正文行文（段数相同但被压缩或多出内容） - 13 个译本：§「原生多模态处理」丢掉 ViT 把图像切块、与文本词向量共存 于同一嵌入空间、自注意力同等对待两类 token 的机制说明；§「本章小结」 丢掉两条分发渠道扩大信任边界因而必须审查描述与版本、隔离凭证的结论 - 11 个译本：§「输出格式要标准化」丢掉「保证子 Agent 考虑所有需要考虑 的方面」这一条理由 - 7 个译本：§「预检-确认两段式」多出「服务端往往不受你控制」等中文没有 的句子；5 个译本 §「动态加载与 KV Cache」多出整段旧稿（各家 API 的 逐条罗列、对模型能力的额外要求） - 另有 §「Sidecar 与提议者-审核者的分工」（4 个译本多出内容）、 §「反馈循环的建立」（ja/vi 仍是旧稿的后训练/外部学习二分）、 §「执行工具的错误代价」（es 多出「感官/手脚」类比）、 §「Skill Hub」（es/hu/id/ta 丢掉 skills.sh 与 ClawHub 的具体信息）、 §「事件触发工具」（id 丢掉注册/触发两个时刻的说明）、 ar 的参数描述用具体例子、返回值与执行代价说明、调用示例、 「先查工具描述再怀疑模型」、静默参数注入等 6 处 另：§「执行工具」一节中，10 个译本把「拒绝熔断器」一段排到了表 4-2 之后，并调换了它与「让安全检查隐形」的先后，现按中文顺序归位。 校验：13 个译本与中文逐节单元数一致；实验编号与所在小节一致；字符量 比值除 ta 一处（泰米尔文字较密，绝对比 2.06 属正常范围）外无异常； 脚注无悬空定义与未定义引用；check_i18n_consistency.py 与 pytest tests/ 通过；译本无 CJK 混入，zhtw 无简体字。 Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 14 天前 | |
docs(i18n): 修复第五章 13 个译本的散文式浓缩与结构漂移 (#1013) 第五章逐节对照中文原稿，修复三类问题： 一、把被压成一段的内容拆回原有段落 - §1 十三个译本都把「七个工具构成极简工具箱」与「这套工具集是 Coding Agent 特有的基础配置，不同于第四章的五类通用工具分类」 合并成一段，现按中文拆回两段。 - §8[5] 类 Vim 编辑命令：十三个译本都丢了后半段（一次思考里发多条 编辑命令会因行号位移而不友好；Vim 是为「看得见状态、每次只规划一 个简单操作」的人设计的，而模型是长思考后成批复杂动作），已补全。 - §2[0] 七个译本、§2[3] 六个译本、ar §8[3]/[4]/[6]、ar §12[13]、 ar §15[16] 均按中文重写补全。 二、删掉中文没有的内容 - §8「实践建议」小节（11 个译本）、§9[9] 第四条要点「持久会话与隔离 的调和」（13 个译本）、ar §5 结尾多出的总结段、hu §6[16] 关于 Sessionless 的对比段。 三、结构与编号 - ar §3[5] 原是一段旧稿（「知识外化原则的自然推论」），与中文的 「AI-ready 团队」判断完全不同，已按中文重写。 - hu §6[15] 把持久化终端会话写成了 PTY / 线程安全缓冲 / 资源泄漏 等中文没有的内容，按中文重写。 - ar §4 补回「代码编写天然处于该象限核心」一段；ta §15 补回「声明式 方法适用于标准化交互场景」一段；vi §13/§14 补回两处小标题。 - es 的图 5-6、图 5-7 脱出了实验引用块，已并回块内（与其余 12 个 译本一致）。 - hu 最后五个实验仍用旧的「Kísérlet 5-11…5-15」写法且编号重复， 改为与本章一致的「5-12. kísérlet」～「5-16. kísérlet」。 - vi 本章实验词统一为 Thử nghiệm（原有 2 处 Thí nghiệm）。 校验：单元计数差异归零；check_i18n_consistency.py 全绿； pytest 802 passed；实验编号、脚注定义/引用、非中文译本 CJK 泄漏、 zhtw 简体字检测均无异常。 Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 14 天前 | |
docs(i18n): 修复第六章 13 个译本的散文式浓缩、重复段与旧稿残留 (#1014) 一、重复内容 - §7[1] 十二个译本把中文的 [1]+[2]+[3] 三段揉进一段，而 [2]、[3] 又照常存在，等于同一段话讲两遍；已剪回中文 [1] 的范围。 - §7[9] 同样问题（[9] 里重复了 [10] 的 n8n 触发器生态），已剪回。 - §15[0] ar/ru/ta/tr/vi 把 Moshi、Interaction Model、GPT-Live 的内容提前塞进首段，与后面的 [1]、[2]、[4] 重复，已剪回；ta/tr 的该段夹杂大量未翻译英文词，一并按中文重写。 二、旧稿残留 - §22 十三个译本开头保留着两段旧稿（「本章花在语音上的篇幅远多于后面 两个场景……」「这三个场景看似不同……」），中文已改写为一段 「语音把时机轴推到了毫秒级……」，现按中文替换。 - §5 十个译本首段是一句多余的 OpenClaw 概述，[1] 又把中文的 [0]+[1] 合并，[4] 还多出一段中文没有的「类别边界」讨论；已按中文重排为 7 段。 三、被压缩或丢失的内容 - §1 十三个译本都缺末段「这一节里模态没有变化，仍然是文本，变的只有 时机。它是从前五章的回合制世界迈出的第一步。」 - §3 十个译本缺「真正的短板在于：对于内置渠道之外的第三方事件源…… 只能等到下一个 Cron/Heartbeat 周期才可能察觉。」 - §12 十一个译本把 [3]、[5] 合成一段并丢掉「本章不展开排队模型」， 已拆回两段。 - §14 ru 把 [1]、[2] 合成一段；§19 ja/ko/zhtw 把两段压成一句。 - §15[2]、§17[0]、§18[0] ja/ko/zhtw 丢掉了具体例证（快模型先建议 购买、慢模型发现套餐缺功能等），已补全；en §5[2] 亦补全。 校验：单元计数差异归零；实验编号 6-1 ~ 6-13 全部对齐；脚注定义/引用 配对无误；无跨语种字符泄漏；check_i18n_consistency.py 全绿； pytest 802 passed。 Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 14 天前 | |
docs(ch7): 说明 τ²-bench 需自行克隆，而非收在配套仓库中（15 译本同步） (#1054) * docs(ch7): 说明 τ²-bench 需自行克隆，而非收在配套仓库中 第七章「一条评估任务的解剖」称源码「位于仓库的 chapter7/tau2-bench」， 但该路径被 .gitignore 第 54 行排除，仓库里并不存在，读者按书查找会落空 （issue #1050）。 τ²-bench 是 Sierra 的开源项目，本仓库刻意不做 vendoring，克隆命令固定在 chapter7/tau2-bench-eval/README.md 中（含 pin 住的上游 commit）。正文改为 指向该 README，并说明克隆到 chapter7/tau2-bench 之后任务文件的位置。 15 个语种同步。 Fixes #1050 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018iSm7JBWoy87hxSpUkJ49T * docs(ch7): 按作者意见收紧措辞，直接讲怎么拿到任务文件 去掉「并未收入配套仓库」的解释和 chapter7/tau2-bench 这个具体路径，改为 一句话说明来源并直接给出操作：克隆到本地后打开任务文件。15 个语种同步。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018iSm7JBWoy87hxSpUkJ49T --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 6 天前 | |
docs(i18n): 修复第八章 13 个译本的散文式浓缩、重复段与旧稿残留 (#1015) 一、整节仍停留在旧稿 - §14「何时选择 Mid-training、SFT 与 RL」：ar/hu/id/ru/ta/tr/vi 七个 译本还是旧结构（引「三阶段全景」、无表8-4、四条决策顺序被换成另一 套论述），已按中文重写整节（含表8-4 四行与四条决策）。 - §18「看似 On-Policy，为什么仍会被数值误差拖垮」：en/he/es 三个译本 是另一版旧稿（公式 + “three forms of instability”），ja/ko/zhtw 把 中文的 [2]+[3] 压成一段并丢掉四条工程建议，六个译本按中文重写。 二、重复与多出的内容 - §12[5] en/he 把整段连同 §12[4] 的要点重复了两遍（en 达 2244 字符， 中文 169），已按中文重写。 - §12[7] 十二个译本尾部多一段引用已不存在的「三阶段全景」小节。 - §8[22] 十一个译本多出「组合泛化 / 上下文学习机制」等中文没有的两句。 - §9 结尾、§38 开头、§39[1] 各有一段中文没有的段落（九个译本）。 三、被压缩或丢失的内容 - §2[1]（Mid-training 与 SFT 损失函数的差别）、§17[1]（重要性比率 的引入句）九个译本整段缺失。 - §32 七个译本把中文的 [2][3][4] 压成两段，丢掉「$T$ 组逐 token 监督」 「不是从零创造能力」「数值一致性检查」等论证。 - §17[0]、§17[3]、§32[0]、§39[0]、§10[0]、§8[4]、§8[13]、§16[0]、 §0[0] 等在 2~11 个译本中被压缩，已逐条按中文补全。 - es 整体落后一版，另有 §1[2]、§9[4]、§11[3]、§14[5]、§14[6]、 §32[2]、§32[4]、§38[0]、§38[2]（11 条常见陷阱只剩提要）、§38[3]、 §39[1]、§39[2] 等 13 处已重写。 四、其他 - 新写的跨节引用一律改用纯文本节名，不使用可能失效的锚点。 - tr 的分词示例原写作「我/喜欢/编/程 四个 token」，与中文的三个 token 不一致，已更正。 校验：单元计数差异归零；实验编号 8-1 ~ 8-19 全部对齐；脚注定义/引用 配对无误；无跨语种字符泄漏；check_i18n_consistency.py 全绿； pytest 802 passed。 Claude-Session: https://claude.ai/code/session_01B1Zu35aad26ZyQbzyAvBJe Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 14 天前 | |
docs(ch9): 以三层验证为主线重写「从运行轨迹中获得学习信号」（15 语种） (#1059) 原节六段读起来跳跃，issue #1031 反映「有点看不懂在说什么」「缺乏上下文 联系」。加小标题只治标，真正的原因有三个： 1. 通篇没说「学习信号」到底长什么样。图9-2 右半边（任务结果／维度评价／ 证据位置／失败标签／证据不足可拒绝评分）才是本节的产出物，正文却只讲 了图的左半边，产出要求反而藏在实验 9-1 的说明里。 2. 六段里并行着三套互不对齐的分类：按任务分（容易验证的／没有标准答案 的）、按验证器分（结果／过程／质量）、按表9-1 分（前五项底线／后两项 服务质量）。而且「结果 vs 过程」本来就藏在第二段内部，读者读到图9-2 把它们拆成两层时必须回头重解析第二段。 3. 表9-1 的七个维度从未与三层对应，看起来像是又一份新清单。 重写为一条主线：评价一条轨迹＝依次回答「是否办成了／是否以允许的方式办成／ 是否让用户舒服」三个问题，对应三层验证器。原有例子（删测试用例、七天退款 承诺、客服耐心与合规变通）全部保留，改挂到各自的层上；表9-1 显式映射回三层； 结尾补一段说明验证器输出的四项要求，把图9-2 右半边接进正文，并接上下一节的 四种更新方法。图9-2 前移到三层展开之前充当地图。 实验 9-1 未改：实验框按设计可独立阅读，与正文有重叠属预期。 顺带修掉 book-ar 表9-1 与其后段落之间缺失的空行（原本会被吸进表格）。 中文版与英文版定稿后，14 个译本同步。ru 与 ta 译本沿用各自现有的章节／图 编号（ru 作「рисунке 8-2」「шестой главе」，ta 作「ஆறாம் அத்தியாயம்」）， 这两处编号偏移是既有问题，留待单独修。 Fixes #1031 Claude-Session: https://claude.ai/code/session_018iSm7JBWoy87hxSpUkJ49T Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 5 天前 | |
docs(book): reorganize chapters for version 2.0 (#909) | 23 天前 | |
docs(ja): add Japanese locale (#84) * docs: add Japanese (ja) locale - mkdocs.yml / lang-switcher.js / build_site.sh: register ja (book-ja/, suffix .ja) - docs/ja/README.md, docs/ja/LEARNING.md + root move-stubs (README.ja.md, docs/LEARNING.ja.md) - Update language switchers, e-book lists and language counts to 6 across all main READMEs and index.md - chapter1-10/README.ja.md (project counts aligned with the Chinese base) - book-ja/: introduction and afterword translated; figures reused from book-en check_i18n_consistency.py passes (6/6). build_epub.sh / build-latest.yml wiring deferred until the 10 chapter bodies and PDF pipeline are ready. * docs(ja): translate chapter bodies (chapter1-10) Translate the full text of all 10 chapters from the Chinese source into book-ja/chapter1.ja.md … chapter10.ja.md (敬体). Structure verified against the source: image references, footnote definitions, headings, tables, code blocks and ★ difficulty ratings preserved; captions rendered as 図X-Y; 实验→実験, 思考题→演習問題; acronyms/product/model names kept in original form. * docs(ja): localize script-generated figures (ch1-5, 8) Add Japanese figure-generation scripts (svg_lib.py + gen_ch{1,2,3,4,5,8}_figs.py, adapted from book-en with display text translated to Japanese) and regenerate their SVGs under book-ja/images/. Code identifiers, coordinates, colors, model/ product names and acronyms are kept verbatim; only human-readable labels, titles, captions and annotations are translated. 63 figures localized. Hand-authored SVGs without a generator (ch6/7/9/10 and a few in ch1-3) still carry English text and will be localized separately. * docs(ja): localize hand-authored figures (ch0/1/2/3/6/7/9/10) Translate the display text of the hand-authored SVG figures that have no generator script (chapter maps, ch6/7/9/10 diagrams, and the remaining figures in ch1-3) from English to Japanese, editing the <text>/<tspan> content in place. Attributes, coordinates, structure, code/tool identifiers, formulas, model/product names and acronyms are preserved. All 135 book-ja figures now render Japanese text and parse as well-formed XML. * docs(ja): add PDF/EPUB build assets for book-ja (build_epub ja case) Add book-ja LaTeX/build assets adapted from book-zhtw: - build_pdf.sh (chapters .ja, ja output name, Japanese title) - preamble.tex (CJK fonts → Hiragino / Noto Serif|Sans CJK JP) - cover.tex (Japanese cover title/subtitle) - crossref.lua (図N-M / 第N章 in-text links) and experiment_box.lua (実験 / 演習問題 boxes) - build_epub.sh: add a ja case (usable via ./build_epub.sh ja) NOT wired into build_epub.sh all or .github/workflows/build-latest.yml on purpose: the Japanese PDF/EPUB build is UNVALIDATED (no LaTeX toolchain here) and enabling it in the release CI before a successful xelatex run could break the main build for the other languages. Enable those after validation. * ci(ja): wire Japanese PDF/EPUB into build-latest (non-fatal / WIP) Add the Japanese edition to the rolling-latest build in a way that CANNOT break the release for the other languages while the ja build is still unvalidated: - add book-ja/** to the trigger paths - install Japanese Noto fonts in a continue-on-error step - build the ja PDF and ja EPUB in separate continue-on-error steps (ja is intentionally NOT added to build_epub.sh all) - copy the ja artifacts only if they were produced (|| echo … skipping) Once cd book-ja && bash build_pdf.sh is confirmed to work in a real xelatex environment, these steps can be promoted to fatal. * fix(ja): point chapter links to the .ja.md files docs/ja/README.md and chapter{1..10}/README.ja.md linked to book-ja/chapterN.md, but the Japanese chapter files carry the .ja suffix (book-ja/chapterN.ja.md) — so every "read chapter" link 404'd. Fix them to match the .ja naming (the same pattern book-vi/book-ta already use for their suffixed editions). * docs(ja): list Japanese in EPUB build docs (EPUB.md) * docs(ja): use the Japanese core formula in chapter1 README Match the rest of the ja edition (docs/ja, lang-switcher, book body): Agent = LLM + コンテキスト + ツール (was: LLM + Context + Tools). * fix(ja,pdf): bundle PNG figures and use a full-coverage CJK-JP font Two issues surfaced by the first ja PDF build (which otherwise succeeded — 356 pages): - The 3 raster figures (n8n-workflow.png, fig2-7.png, attention-visualization.png) were skipped by a global *.png gitignore, so pandoc could not fetch them. Force-add them (book-en tracks the same files). - Hiragino Mincho ProN lacks Simplified-Chinese glyphs that legitimately appear in the book (verbatim code comments and Chinese product names like 硅基流动 / SiliconFlow), producing "Missing character" tofu. Prefer Noto Serif/Sans CJK JP (full CJK coverage; confirmed installed on the CI runner), falling back to Hiragino only if Noto is absent. - Override the TOC label 目录 → 目次 (ElegantBook has no lang=jp). The remaining CI failure is only the "Upload" step (the fork has no latest release); it is unrelated to the ja build and does not affect the other languages. * docs(ja): translate reference answers (思考题参考答案.ja.md) Translate the reference answers to the chapters' thinking questions from the Chinese source. Gives ja parity with the zh/zh-TW editions and makes the language switcher's 演習問題の解答例 entry resolve. 敬体; glossary followed; code blocks and acronyms/model names kept verbatim. * fix(ja,pdf): render the — (U+2015) dash via the CJK fallback font The second ja PDF build (363 pages, all Simplified-Chinese and PNG issues gone) left one warning: 812× "Missing character U+2015 (―)". The horizontal bar is used throughout the translation as the Japanese dash 「――」 but is absent from the Latin body font (lmroman); route it to the CJK-capable fallback font (Noto Sans CJK JP), matching how ✓/✗/≈ and other symbols are already handled. --------- Co-authored-by: Bojie Li <bojieli@gmail.com> | 1 个月前 | |
docs(ja): add Japanese locale (#84) * docs: add Japanese (ja) locale - mkdocs.yml / lang-switcher.js / build_site.sh: register ja (book-ja/, suffix .ja) - docs/ja/README.md, docs/ja/LEARNING.md + root move-stubs (README.ja.md, docs/LEARNING.ja.md) - Update language switchers, e-book lists and language counts to 6 across all main READMEs and index.md - chapter1-10/README.ja.md (project counts aligned with the Chinese base) - book-ja/: introduction and afterword translated; figures reused from book-en check_i18n_consistency.py passes (6/6). build_epub.sh / build-latest.yml wiring deferred until the 10 chapter bodies and PDF pipeline are ready. * docs(ja): translate chapter bodies (chapter1-10) Translate the full text of all 10 chapters from the Chinese source into book-ja/chapter1.ja.md … chapter10.ja.md (敬体). Structure verified against the source: image references, footnote definitions, headings, tables, code blocks and ★ difficulty ratings preserved; captions rendered as 図X-Y; 实验→実験, 思考题→演習問題; acronyms/product/model names kept in original form. * docs(ja): localize script-generated figures (ch1-5, 8) Add Japanese figure-generation scripts (svg_lib.py + gen_ch{1,2,3,4,5,8}_figs.py, adapted from book-en with display text translated to Japanese) and regenerate their SVGs under book-ja/images/. Code identifiers, coordinates, colors, model/ product names and acronyms are kept verbatim; only human-readable labels, titles, captions and annotations are translated. 63 figures localized. Hand-authored SVGs without a generator (ch6/7/9/10 and a few in ch1-3) still carry English text and will be localized separately. * docs(ja): localize hand-authored figures (ch0/1/2/3/6/7/9/10) Translate the display text of the hand-authored SVG figures that have no generator script (chapter maps, ch6/7/9/10 diagrams, and the remaining figures in ch1-3) from English to Japanese, editing the <text>/<tspan> content in place. Attributes, coordinates, structure, code/tool identifiers, formulas, model/product names and acronyms are preserved. All 135 book-ja figures now render Japanese text and parse as well-formed XML. * docs(ja): add PDF/EPUB build assets for book-ja (build_epub ja case) Add book-ja LaTeX/build assets adapted from book-zhtw: - build_pdf.sh (chapters .ja, ja output name, Japanese title) - preamble.tex (CJK fonts → Hiragino / Noto Serif|Sans CJK JP) - cover.tex (Japanese cover title/subtitle) - crossref.lua (図N-M / 第N章 in-text links) and experiment_box.lua (実験 / 演習問題 boxes) - build_epub.sh: add a ja case (usable via ./build_epub.sh ja) NOT wired into build_epub.sh all or .github/workflows/build-latest.yml on purpose: the Japanese PDF/EPUB build is UNVALIDATED (no LaTeX toolchain here) and enabling it in the release CI before a successful xelatex run could break the main build for the other languages. Enable those after validation. * ci(ja): wire Japanese PDF/EPUB into build-latest (non-fatal / WIP) Add the Japanese edition to the rolling-latest build in a way that CANNOT break the release for the other languages while the ja build is still unvalidated: - add book-ja/** to the trigger paths - install Japanese Noto fonts in a continue-on-error step - build the ja PDF and ja EPUB in separate continue-on-error steps (ja is intentionally NOT added to build_epub.sh all) - copy the ja artifacts only if they were produced (|| echo … skipping) Once cd book-ja && bash build_pdf.sh is confirmed to work in a real xelatex environment, these steps can be promoted to fatal. * fix(ja): point chapter links to the .ja.md files docs/ja/README.md and chapter{1..10}/README.ja.md linked to book-ja/chapterN.md, but the Japanese chapter files carry the .ja suffix (book-ja/chapterN.ja.md) — so every "read chapter" link 404'd. Fix them to match the .ja naming (the same pattern book-vi/book-ta already use for their suffixed editions). * docs(ja): list Japanese in EPUB build docs (EPUB.md) * docs(ja): use the Japanese core formula in chapter1 README Match the rest of the ja edition (docs/ja, lang-switcher, book body): Agent = LLM + コンテキスト + ツール (was: LLM + Context + Tools). * fix(ja,pdf): bundle PNG figures and use a full-coverage CJK-JP font Two issues surfaced by the first ja PDF build (which otherwise succeeded — 356 pages): - The 3 raster figures (n8n-workflow.png, fig2-7.png, attention-visualization.png) were skipped by a global *.png gitignore, so pandoc could not fetch them. Force-add them (book-en tracks the same files). - Hiragino Mincho ProN lacks Simplified-Chinese glyphs that legitimately appear in the book (verbatim code comments and Chinese product names like 硅基流动 / SiliconFlow), producing "Missing character" tofu. Prefer Noto Serif/Sans CJK JP (full CJK coverage; confirmed installed on the CI runner), falling back to Hiragino only if Noto is absent. - Override the TOC label 目录 → 目次 (ElegantBook has no lang=jp). The remaining CI failure is only the "Upload" step (the fork has no latest release); it is unrelated to the ja build and does not affect the other languages. * docs(ja): translate reference answers (思考题参考答案.ja.md) Translate the reference answers to the chapters' thinking questions from the Chinese source. Gives ja parity with the zh/zh-TW editions and makes the language switcher's 演習問題の解答例 entry resolve. 敬体; glossary followed; code blocks and acronyms/model names kept verbatim. * fix(ja,pdf): render the — (U+2015) dash via the CJK fallback font The second ja PDF build (363 pages, all Simplified-Chinese and PNG issues gone) left one warning: 812× "Missing character U+2015 (―)". The horizontal bar is used throughout the translation as the Japanese dash 「――」 but is absent from the Latin body font (lmroman); route it to the CJK-capable fallback font (Noto Sans CJK JP), matching how ✓/✗/≈ and other symbols are already handled. --------- Co-authored-by: Bojie Li <bojieli@gmail.com> | 1 个月前 | |
docs(i18n): tighten chapter transitions across editions (#912) | 23 天前 | |
Fix page numbers on chapter opening pages | 1 个月前 | |
docs(i18n): 把第五章「接管」一节与实验 5-1、5-2 同步到 13 个语种 (#976) - 13 个语种的第五章「故障与错误恢复」补上接管一段（思考由可读文字与厂商 凭证两部分构成、能带走的是文字带不走的是凭证、按最严格的一端设计并准备 文本化退路、轨迹存中立格式），以及实验 5-1、5-2 两个实验框 - 各语种原 5-1~5-14 顺延为 5-3~5-16，本章小结的「检测、恢复与终止」改为 四项，reference-answers 的交叉引用同步 - 各语种第二章思维链回传一段补一句指向实验 5-1 - 逐语种核对位移严格等于「main + 2」：越南语两个实验框用的是 Thí nghiệm 而非 Thử nghiệm，此前被漏掉，已一并补齐 沿用各语种既有的实验框标签与编号写法（匈牙利语的「5-N. kísérlet」、 希伯来语的不换行连字符等）。图号、表号、时长区间等非实验编号未改动。 正文见已合入的 #974，实验实现与证据见 #975 Claude-Session: https://claude.ai/code/session_015ni6QYiSBGTPbFJ9Nm819x Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> | 15 天前 |
| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
| 12 天前 | ||
| 25 天前 | ||
| 23 天前 | ||
| 9 天前 | ||
| 9 天前 | ||
| 6 天前 | ||
| 14 天前 | ||
| 14 天前 | ||
| 14 天前 | ||
| 14 天前 | ||
| 6 天前 | ||
| 14 天前 | ||
| 5 天前 | ||
| 23 天前 | ||
| 1 个月前 | ||
| 1 个月前 | ||
| 23 天前 | ||
| 1 个月前 | ||
| 15 天前 |