{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"真实猫图结果：保留模型、照片与历史身份","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.md","pages":[{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","unit":"已知图文配对怎样变成可用的文字候选接口？","unitId":"clip-pairing-to-candidates","chapter":"III","number":1}],"aliases":[{"oldId":"b-b6feab44239c3d7f439ac337","targetId":"b-b6feab44239c3d7f439ac337","page":"reading-opening","operation":"move","beforeSHA256":"e0a10b9509c9dcebcdb8dd523f9126e067c84af9d61fc5636ee52c059738fbad","afterSHA256":"e0a10b9509c9dcebcdb8dd523f9126e067c84af9d61fc5636ee52c059738fbad","reason":"Same source body in authored semantic page"},{"oldId":"b-d955922ad336a1e175fe5fc2","targetId":"b-d955922ad336a1e175fe5fc2","page":"reading-opening","operation":"move","beforeSHA256":"b9a790420611560584ad9070f7e9e60c9da1afe63b7e0688164e330d859589ab","afterSHA256":"b9a790420611560584ad9070f7e9e60c9da1afe63b7e0688164e330d859589ab","reason":"Same source body in authored semantic page"},{"oldId":"b-045b6fed684f8d7e73bc55a1","targetId":"b-045b6fed684f8d7e73bc55a1","page":"reading-opening","operation":"move","beforeSHA256":"08592cac5e85947afd40374d30d6030489704a2787422ccca7752470cb945f62","afterSHA256":"08592cac5e85947afd40374d30d6030489704a2787422ccca7752470cb945f62","reason":"Same source body in authored semantic page"},{"oldId":"b-38a892a9c041369febb7affc","targetId":"b-38a892a9c041369febb7affc","page":"reading-opening","operation":"move","beforeSHA256":"c7061a8ad0794772fc5fd4b5f0bdb6e66e098c0aaa2732e4c041ff8bb739810e","afterSHA256":"c7061a8ad0794772fc5fd4b5f0bdb6e66e098c0aaa2732e4c041ff8bb739810e","reason":"Same source body in authored semantic page"},{"oldId":"b-9bcd7695525f8ffafe9770a5","targetId":"b-9bcd7695525f8ffafe9770a5","page":"reading-opening","operation":"move","beforeSHA256":"1fb9bd54f6a6d811bd1b68919726fd1177e6a5264ddb1afa54180876084827ba","afterSHA256":"1fb9bd54f6a6d811bd1b68919726fd1177e6a5264ddb1afa54180876084827ba","reason":"Same source body in authored semantic page"},{"oldId":"b-975e9d4fc2b5229e5c173a6f","targetId":"b-975e9d4fc2b5229e5c173a6f","page":"reading-opening","operation":"move","beforeSHA256":"3fd55da4e643785a3ea3b5949b03d73e798ccf3ae1b5812997a45c182c9d7898","afterSHA256":"3fd55da4e643785a3ea3b5949b03d73e798ccf3ae1b5812997a45c182c9d7898","reason":"Same source body in authored semantic page"},{"oldId":"b-2a678d28a52315a3f2ce28f7","targetId":"b-2a678d28a52315a3f2ce28f7","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-b6feab44239c3d7f439ac337","type":"h1","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-b6feab44239c3d7f439ac337","content":"# 真实猫图结果：保留模型、照片与历史身份","sha256":"e0a10b9509c9dcebcdb8dd523f9126e067c84af9d61fc5636ee52c059738fbad","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-d955922ad336a1e175fe5fc2","prerequisites":[{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d955922ad336a1e175fe5fc2","type":"p","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-d955922ad336a1e175fe5fc2","content":"这里的 zero-shot 不使用当前下游数据集再训练模型，而用已有编码器比较文字候选。增加候选和更换模板是不同干预：前者可以只变 softmax 分母，后者会改变文本向量。LAB04 的真实猫图四臂比较保留，可直接阅读已有结果，不必为理解接口重跑模型。","sha256":"b9a790420611560584ad9070f7e9e60c9da1afe63b7e0688164e330d859589ab","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b6feab44239c3d7f439ac337","nextBlock":"b-045b6fed684f8d7e73bc55a1","prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"}],"relatedBlocks":[]},{"id":"b-045b6fed684f8d7e73bc55a1","type":"p","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-045b6fed684f8d7e73bc55a1","content":"当前 LAB04 保存真实 Chelsea 猫照片与 OpenAI CLIP ViT-B/32 实验。2026-09-22 的本地复核使用固定 checkpoint revision `3d74acf9a28c67741b2f4f2ea7635f0aaf6f0268`，没有重新训练 CLIP；当前整理也没有重跑该推理。图片先转 RGB、缩放并中心裁剪到 $224\\times224$，再按模型配置归一化。","sha256":"08592cac5e85947afd40374d30d6030489704a2787422ccca7752470cb945f62","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"224\\times224","definitionContext":"真实猫图结果：保留模型、照片与历史身份"}],"previousBlock":"b-d955922ad336a1e175fe5fc2","nextBlock":"b-38a892a9c041369febb7affc","prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"},{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-38a892a9c041369febb7affc","type":"p","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-38a892a9c041369febb7affc","content":"基础 cat/dog/horse 候选下，cat 余弦约 0.28691、候选 softmax 约 0.99577；添加 tiger 与 fox 后，共享 cat 的余弦保持不变，softmax 降至约 0.98195；再删除 cat，tiger 成为剩余候选的最高项。新旧猫图余弦最大差约 $1.04\\times10^{-7}$，把后来的复核与历史材料连接起来。","sha256":"c7061a8ad0794772fc5fd4b5f0bdb6e66e098c0aaa2732e4c041ff8bb739810e","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"1.04\\times10^{-7}","definitionContext":"真实猫图结果：保留模型、照片与历史身份"}],"previousBlock":"b-045b6fed684f8d7e73bc55a1","nextBlock":"b-9bcd7695525f8ffafe9770a5","prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"},{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9bcd7695525f8ffafe9770a5","type":"p","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-9bcd7695525f8ffafe9770a5","content":"正面结果是模型可以在这张真实照片上把图像与合理文字对应，并且其候选接口的变化符合预期。数值 0.99577 是给定候选集合内的相对分配，不是“现实中有猫的校准概率”。一张猫图也不能替代数据集级泛化评测。","sha256":"1fb9bd54f6a6d811bd1b68919726fd1177e6a5264ddb1afa54180876084827ba","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-38a892a9c041369febb7affc","nextBlock":"b-975e9d4fc2b5229e5c173a6f","prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"},{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-975e9d4fc2b5229e5c173a6f","type":"p","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-975e9d4fc2b5229e5c173a6f","content":"LAB04 负责模型输入、候选、模板和关系检查；LAB12 复用同一模型产生的完整 512 维向量，研究图库缓存与搜索。不要把四个候选分数拼成所谓图像表示；真正可复用的是编码器输出的向量。先在 LAB04 理解分数身份，再在 LAB12 看系统组织，就无需重复训练或建立两份同义实验。","sha256":"3fd55da4e643785a3ea3b5949b03d73e798ccf3ae1b5812997a45c182c9d7898","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9bcd7695525f8ffafe9770a5","nextBlock":"b-2a678d28a52315a3f2ce28f7","prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"},{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2a678d28a52315a3f2ce28f7","type":"h2","scope":"clip-real-checkpoint-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-real-checkpoint-evidence.html#b-2a678d28a52315a3f2ce28f7","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-pairing-to-candidates","title":"已知图文配对怎样变成可用的文字候选接口？"},"page":{"id":"reading-opening","title":"真实猫图结果：保留模型、照片与历史身份","number":1},"sectionTitle":"真实猫图结果：保留模型、照片与历史身份","currentQuestion":"真实猫图结果：保留模型、照片与历史身份","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-975e9d4fc2b5229e5c173a6f","nextBlock":null,"prerequisites":[{"id":"b-b6feab44239c3d7f439ac337","url":"#b-b6feab44239c3d7f439ac337","role":"section-title"},{"id":"b-d955922ad336a1e175fe5fc2","url":"#b-d955922ad336a1e175fe5fc2","role":"problem-context"}],"relatedBlocks":[]}]}