{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"检索为什么能缓存：每张图编码一次，查询再比较","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.md","pages":[{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","unit":"检索到底复用了什么，又省去了什么？","unitId":"retrieval-cache-and-evaluation","chapter":"III","number":1}],"aliases":[{"oldId":"b-7c842ded1098a18239e9968d","targetId":"b-7c842ded1098a18239e9968d","page":"reading-opening","operation":"move","beforeSHA256":"af155c3801db46c8a49d2e3b97157107dbeeef9e8a96b754e9ef8869bf0360de","afterSHA256":"af155c3801db46c8a49d2e3b97157107dbeeef9e8a96b754e9ef8869bf0360de","reason":"Same source body in authored semantic page"},{"oldId":"b-f94d48b6db527affc11defc0","targetId":"b-f94d48b6db527affc11defc0","page":"reading-opening","operation":"move","beforeSHA256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","afterSHA256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","reason":"Same source body in authored semantic page"},{"oldId":"b-3b54546be999277828c2ef7c","targetId":"b-3b54546be999277828c2ef7c","page":"reading-opening","operation":"move","beforeSHA256":"0d52d45b91d64683943b3870efe42eb2211cb249baaa5d700568f83f16ac824f","afterSHA256":"0d52d45b91d64683943b3870efe42eb2211cb249baaa5d700568f83f16ac824f","reason":"Same source body in authored semantic page"},{"oldId":"b-2e26938ea192ee911379e72e","targetId":"b-2e26938ea192ee911379e72e","page":"reading-opening","operation":"move","beforeSHA256":"548903116868e99139fe6cb659aa140a85b965dcc46a1ad6a79904bd982e3beb","afterSHA256":"548903116868e99139fe6cb659aa140a85b965dcc46a1ad6a79904bd982e3beb","reason":"Same source body in authored semantic page"},{"oldId":"b-e6aaa594223a51a4d9b3999b","targetId":"b-e6aaa594223a51a4d9b3999b","page":"reading-opening","operation":"move","beforeSHA256":"a0b9c4c8d9eebe422e20a0d9595195dc58efc43923716e264bd41dc0bfb60c5d","afterSHA256":"a0b9c4c8d9eebe422e20a0d9595195dc58efc43923716e264bd41dc0bfb60c5d","reason":"Same source body in authored semantic page"},{"oldId":"b-c72dbf2647c42d8ec4037148","targetId":"b-c72dbf2647c42d8ec4037148","page":"reading-opening","operation":"move","beforeSHA256":"92827721d51ccba74677884e288a5b789ab5ded7a0520fdc0992941218d8ee81","afterSHA256":"92827721d51ccba74677884e288a5b789ab5ded7a0520fdc0992941218d8ee81","reason":"Same source body in authored semantic page"},{"oldId":"b-f98faa11966d02406e93c0cb","targetId":"b-f98faa11966d02406e93c0cb","page":"reading-opening","operation":"rewrite","beforeSHA256":"0177d59718715cd38702da621c16065c630e29d66e34077d29389bb96f1d6b6e","afterSHA256":"043e052363a715dc27a99d9b09d9147a12bb987e2132f4e6a8528adf72775fc7","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-0aaf2f07a3cf9e3a3692b6cb","targetId":"b-0aaf2f07a3cf9e3a3692b6cb","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-7c842ded1098a18239e9968d","type":"h1","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-7c842ded1098a18239e9968d","content":"# 检索为什么能缓存：每张图编码一次，查询再比较","sha256":"af155c3801db46c8a49d2e3b97157107dbeeef9e8a96b754e9ef8869bf0360de","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-f94d48b6db527affc11defc0","prerequisites":[{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-f94d48b6db527affc11defc0","type":"p","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-f94d48b6db527affc11defc0","content":"检索把同一个比较空间用于一组已有图片。图库向量可复用，查询只需编码一次再搜索；页面前三张结果来自已有真实缓存。工作量、相关性指标和近似索引在说明与 LAB12 中展开；检索系统的进一步组织留待第八讲。","sha256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7c842ded1098a18239e9968d","nextBlock":"b-3b54546be999277828c2ef7c","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-3b54546be999277828c2ef7c","type":"p","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-3b54546be999277828c2ef7c","content":"设图库有 $N$ 张图，收到 $Q$ 条文字查询。双编码器的图像输出不依赖当前查询，所以可以先计算图库矩阵 $V\\in\\mathbb R^{N\\times d}$，查询到来时只计算文字向量 $u$，再用 $Vu$ 得分。这把“理解输入的深度编码”和“在现有表示中比较”分成两段。","sha256":"0d52d45b91d64683943b3870efe42eb2211cb249baaa5d700568f83f16ac824f","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"N","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"Q","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"V\\in\\mathbb R^{N\\times d}","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"u","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"Vu","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"}],"previousBlock":"b-f94d48b6db527affc11defc0","nextBlock":"b-2e26938ea192ee911379e72e","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-2e26938ea192ee911379e72e","type":"p","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-2e26938ea192ee911379e72e","content":"若每个图文对都重新分别编码，需 $NQ$ 次图片样本编码和 $NQ$ 次文字样本编码；有缓存则是 $N+Q$。本课程 $N=24,Q=12$，两者为 576 与 36 次样本编码。它是解析工作量，不是秒数；一个批次调用可以同时编码多个样本。","sha256":"548903116868e99139fe6cb659aa140a85b965dcc46a1ad6a79904bd982e3beb","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"NQ","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"NQ","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"N+Q","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"N=24,Q=12","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"}],"previousBlock":"b-3b54546be999277828c2ef7c","nextBlock":"b-e6aaa594223a51a4d9b3999b","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-e6aaa594223a51a4d9b3999b","type":"p","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-e6aaa594223a51a4d9b3999b","content":"缓存并没有消除搜索本身。exact 查询仍要与全部 $N$ 向量做点积；float32 图库矩阵占 $4Nd$ 字节。本例 $24\\times512\\times4=49,152$ 字节，不含图片、模型、ID 与索引。改变 checkpoint、预处理或图片内容需要重建对应缓存；只新增文字候选不改变旧图片向量。","sha256":"a0b9c4c8d9eebe422e20a0d9595195dc58efc43923716e264bd41dc0bfb60c5d","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"N","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"4Nd","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"},{"latex":"24\\times512\\times4=49,152","definitionContext":"检索为什么能缓存：每张图编码一次，查询再比较"}],"previousBlock":"b-2e26938ea192ee911379e72e","nextBlock":"b-c72dbf2647c42d8ec4037148","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-c72dbf2647c42d8ec4037148","type":"p","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-c72dbf2647c42d8ec4037148","content":"P59 使用已有 24 项缓存显示猫查询的前三项：`chelsea-center90`、`chelsea-full`、`chelsea-right90`。24 项其实来自六张照片的四种相关裁剪，不能当成 24 张独立语义样本。它足以观察排名与缓存复用；更大的索引工程属于第八讲和 LAB12 的选读部分。","sha256":"92827721d51ccba74677884e288a5b789ab5ded7a0520fdc0992941218d8ee81","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e6aaa594223a51a4d9b3999b","nextBlock":"b-f98faa11966d02406e93c0cb","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]},{"id":"b-f98faa11966d02406e93c0cb","type":"figure","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-f98faa11966d02406e93c0cb","content":"图内标签与数值：6 张来源照片 × 4 个相关视图 = 24 项；查询：“a photo of a cat.”；文字编码一次；与 24 个已保存图像向量比较；1   chelsea-center90；2   chelsea-full；3   chelsea-right90；已有结果：精确检索的前三项都是猫图变体；复用 2026-09-22 本地真实 CLIP 缓存；24 项含相关裁剪，不是 24 个独立来源。本轮未运行模型。\n\n查询对应的前三张结果来自已经保存的真实 CLIP 缓存。 查看对应图解状态\n\n- 6 张来源照片 × 4 个相关视图 = 24 项\n- 查询：“a photo of a cat.”\n- 文字编码一次\n- 与 24 个已保存图像向量比较\n- 1 chelsea-center90\n- 2 chelsea-full\n- 3 chelsea-right90\n- 已有结果：精确检索的前三项都是猫图变体\n- 复用 2026-09-22 本地真实 CLIP 缓存；24 项含相关裁剪，不是 24 个独立来源。本轮未运行模型。查询对应的前三张结果来自已经保存的真实 CLIP 缓存。 [查看对应图解状态](/#/l3-r2-p59/0)[查看完整图解](/#/l3-r2-p59/0)","sha256":"043e052363a715dc27a99d9b09d9147a12bb987e2132f4e6a8528adf72775fc7","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"查询对应的前三张结果来自已经保存的真实 CLIP 缓存。 查看对应图解状态","labels":["6 张来源照片 × 4 个相关视图 = 24 项","查询：“a photo of a cat.”","文字编码一次","与 24 个已保存图像向量比较","1   chelsea-center90","2   chelsea-full","3   chelsea-right90","已有结果：精确检索的前三项都是猫图变体","复用 2026-09-22 本地真实 CLIP 缓存；24 项含相关裁剪，不是 24 个独立来源。本轮未运行模型。"],"description":"查询对应的前三张结果来自已经保存的真实 CLIP 缓存。 查看对应图解状态6 张来源照片 × 4 个相关视图 = 24 项查询：“a photo of a cat.”文字编码一次与 24 个已保存图像向量比较1 chelsea-center902 chelsea-full3 chelsea-right90已有结果：精确检索的前三项都是猫图变体复用 2026-09-22 本地真实 CLIP 缓存；24 项含相关裁剪，不是 24 个独立来源。本轮未运行模型。","evidenceIdentity":"查询对应的前三张结果来自已经保存的真实 CLIP 缓存。 查看对应图解状态","images":[{"src":"/assets/display/61b4b65c7be015d923cad24b.webp","alt":""}]},"previousBlock":"b-c72dbf2647c42d8ec4037148","nextBlock":"b-0aaf2f07a3cf9e3a3692b6cb","prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0aaf2f07a3cf9e3a3692b6cb","type":"h2","scope":"retrieval-cache-interface","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-cache-interface.html#b-0aaf2f07a3cf9e3a3692b6cb","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"检索为什么能缓存：每张图编码一次，查询再比较","number":1},"sectionTitle":"检索为什么能缓存：每张图编码一次，查询再比较","currentQuestion":"检索为什么能缓存：每张图编码一次，查询再比较","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f98faa11966d02406e93c0cb","nextBlock":null,"prerequisites":[{"id":"b-7c842ded1098a18239e9968d","url":"#b-7c842ded1098a18239e9968d","role":"section-title"},{"id":"b-f94d48b6db527affc11defc0","url":"#b-f94d48b6db527affc11defc0","role":"problem-context"}],"relatedBlocks":[{"id":"b-f98faa11966d02406e93c0cb","url":"#b-f98faa11966d02406e93c0cb","relation":"figure-and-explanation"}]}]}
