{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"两个编码器，一个配对目标","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.md","pages":[{"id":"reading-opening","title":"两个编码器，一个配对目标","unit":"两个编码器，一个配对目标","unitId":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","chapter":"course","number":1}],"aliases":[{"oldId":"b-c8d9497113fedff65cab6359","targetId":"b-c8d9497113fedff65cab6359","page":"reading-opening","operation":"move","beforeSHA256":"fd47961b052e3272eb8f0b09b2580de4f8630f510ef0f0dda7e4b1b6d97f84bf","afterSHA256":"fd47961b052e3272eb8f0b09b2580de4f8630f510ef0f0dda7e4b1b6d97f84bf","reason":"Same source body in authored semantic page"},{"oldId":"b-2369c3f277c9225590898fa2","targetId":"b-2369c3f277c9225590898fa2","page":"reading-opening","operation":"move","beforeSHA256":"7451a2bbfbb520df8abe4a0893e7fb0596aa1bdb233913da734b891cbc8c875b","afterSHA256":"7451a2bbfbb520df8abe4a0893e7fb0596aa1bdb233913da734b891cbc8c875b","reason":"Same source body in authored semantic page"},{"oldId":"b-abc9aa30a017cb9017a93242","targetId":"b-abc9aa30a017cb9017a93242","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-35f8b93fae4b4187863759c1","targetId":"b-35f8b93fae4b4187863759c1","page":"reading-opening","operation":"move","beforeSHA256":"8b4a38fbe2d21e8c471d155a892cffc7decd7adfc064650236630fb3edf7a7d0","afterSHA256":"8b4a38fbe2d21e8c471d155a892cffc7decd7adfc064650236630fb3edf7a7d0","reason":"Same source body in authored semantic page"},{"oldId":"b-2b0d51625875cdb99f08bd08","targetId":"b-2b0d51625875cdb99f08bd08","page":"reading-opening","operation":"move","beforeSHA256":"bab4bd804571aa5afbaf197e9c1539510fc766578c9825aa46cadbb18444cc32","afterSHA256":"bab4bd804571aa5afbaf197e9c1539510fc766578c9825aa46cadbb18444cc32","reason":"Same source body in authored semantic page"},{"oldId":"b-37fbfe15a840fa86449d8fd0","targetId":"b-37fbfe15a840fa86449d8fd0","page":"reading-opening","operation":"move","beforeSHA256":"50590b614b01aacf740b6f59f0a25be53dce523d2843825124cc6e1195f0f09c","afterSHA256":"50590b614b01aacf740b6f59f0a25be53dce523d2843825124cc6e1195f0f09c","reason":"Same source body in authored semantic page"},{"oldId":"b-87a343db0d5cf1dac70e6445","targetId":"b-87a343db0d5cf1dac70e6445","page":"reading-opening","operation":"move","beforeSHA256":"2e7f9ca1ab75355dcb74d4a3f8cd95d8ec9fc0ddc39e8919ee0ca1d8fa2e2052","afterSHA256":"2e7f9ca1ab75355dcb74d4a3f8cd95d8ec9fc0ddc39e8919ee0ca1d8fa2e2052","reason":"Same source body in authored semantic page"},{"oldId":"b-53328cdd6d3c71b56eebfe62","targetId":"b-53328cdd6d3c71b56eebfe62","page":"reading-opening","operation":"move","beforeSHA256":"2b31d6a9ce476da9fbc54ff1cd84e4d23c0a209eef12ce6ab7054f0dfa6f5c76","afterSHA256":"2b31d6a9ce476da9fbc54ff1cd84e4d23c0a209eef12ce6ab7054f0dfa6f5c76","reason":"Same source body in authored semantic page"},{"oldId":"b-ef2dd01ce398079aa5cbe781","targetId":"b-ef2dd01ce398079aa5cbe781","page":"reading-opening","operation":"move","beforeSHA256":"bfed64de92e5326d9f2ea08a5fbf216a9101f67d42e71710906c6ab6d2692320","afterSHA256":"bfed64de92e5326d9f2ea08a5fbf216a9101f67d42e71710906c6ab6d2692320","reason":"Same source body in authored semantic page"},{"oldId":"b-011878768e3079d2d392bcd5","targetId":"b-011878768e3079d2d392bcd5","page":"reading-opening","operation":"move","beforeSHA256":"063457994938d4e5dfe61567141bd163cb7d7cce6959928c1c0eb5510469ab6b","afterSHA256":"063457994938d4e5dfe61567141bd163cb7d7cce6959928c1c0eb5510469ab6b","reason":"Same source body in authored semantic page"},{"oldId":"b-5f730bbd8eff7ad501b1df09","targetId":"b-5f730bbd8eff7ad501b1df09","page":"reading-opening","operation":"move","beforeSHA256":"421bb86160826d098ebd40ff162672a976ef2d422bd0fce72fa666c4eaad7d3e","afterSHA256":"421bb86160826d098ebd40ff162672a976ef2d422bd0fce72fa666c4eaad7d3e","reason":"Same source body in authored semantic page"},{"oldId":"b-3a497f200caba71b2da46c8e","targetId":"b-3a497f200caba71b2da46c8e","page":"reading-opening","operation":"move","beforeSHA256":"eba1b58a7fce19f59185999c38f5f654b59946a33ae131684e1abbc607f712e6","afterSHA256":"eba1b58a7fce19f59185999c38f5f654b59946a33ae131684e1abbc607f712e6","reason":"Same source body in authored semantic page"},{"oldId":"b-9833912d287f42801ed86c98","targetId":"b-9833912d287f42801ed86c98","page":"reading-opening","operation":"move","beforeSHA256":"6b5dab1b128812a45c0237b9943cd89a557634629a8386260132bba6eb7d7a14","afterSHA256":"6b5dab1b128812a45c0237b9943cd89a557634629a8386260132bba6eb7d7a14","reason":"Same source body in authored semantic page"},{"oldId":"b-4757d5a416e5d9d398b7ea86","targetId":"b-4757d5a416e5d9d398b7ea86","page":"reading-opening","operation":"move","beforeSHA256":"b82eb99f0b75c3c6d07cae11fdb8041ed3871643065e2b41ebf2f1d956f99de4","afterSHA256":"b82eb99f0b75c3c6d07cae11fdb8041ed3871643065e2b41ebf2f1d956f99de4","reason":"Same source body in authored semantic page"},{"oldId":"b-20adc2b98c5178f1be41aa69","targetId":"b-20adc2b98c5178f1be41aa69","page":"reading-opening","operation":"move","beforeSHA256":"eeb43459eb9f075a95c7403ff26d52e85c14cf632e9eef7d81193d82eb7eae86","afterSHA256":"eeb43459eb9f075a95c7403ff26d52e85c14cf632e9eef7d81193d82eb7eae86","reason":"Same source body in authored semantic page"},{"oldId":"b-c857dae38be3a3c7e43ae650","targetId":"b-c857dae38be3a3c7e43ae650","page":"reading-opening","operation":"move","beforeSHA256":"e2fd6ed94ff5812ee027b0c1d2a3d46127eb13403d0e3f6c267ac4ef37f49552","afterSHA256":"e2fd6ed94ff5812ee027b0c1d2a3d46127eb13403d0e3f6c267ac4ef37f49552","reason":"Same source body in authored semantic page"},{"oldId":"b-70c737fe2d1e5f5c4a362267","targetId":"b-70c737fe2d1e5f5c4a362267","page":"reading-opening","operation":"move","beforeSHA256":"38c3c7217ff456579df20fe9c5703f47cfc3e63d9dcbb6bd611964eef7aa9ad0","afterSHA256":"38c3c7217ff456579df20fe9c5703f47cfc3e63d9dcbb6bd611964eef7aa9ad0","reason":"Same source body in authored semantic page"},{"oldId":"block-ed31be25c84b4c4eb511e2b846038ac3--两个编码器一个配对目标","targetId":"b-011878768e3079d2d392bcd5","operation":"retain-source-identity"},{"oldId":"block-ed31be25c84b4c4eb511e2b846038ac3--为什么这是可用的新接口","targetId":"b-3a497f200caba71b2da46c8e","operation":"retain-source-identity"},{"oldId":"block-ed31be25c84b4c4eb511e2b846038ac3--三个干预三个不同问题","targetId":"b-4757d5a416e5d9d398b7ea86","operation":"retain-source-identity"}],"blocks":[{"id":"b-c8d9497113fedff65cab6359","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-c8d9497113fedff65cab6359","content":"[CLS 10 · CLIP：配对学习与可复用表示](/course/notebooks/d890081ddce04dac9aec8238b2809250.html)","sha256":"fd47961b052e3272eb8f0b09b2580de4f8630f510ef0f0dda7e4b1b6d97f84bf","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-2369c3f277c9225590898fa2","prerequisites":[],"relatedBlocks":[]},{"id":"b-2369c3f277c9225590898fa2","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-2369c3f277c9225590898fa2","content":"# 两个编码器，一个配对目标","sha256":"7451a2bbfbb520df8abe4a0893e7fb0596aa1bdb233913da734b891cbc8c875b","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c8d9497113fedff65cab6359","nextBlock":"b-abc9aa30a017cb9017a93242","prerequisites":[{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-abc9aa30a017cb9017a93242","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-abc9aa30a017cb9017a93242","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2369c3f277c9225590898fa2","nextBlock":"b-35f8b93fae4b4187863759c1","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-35f8b93fae4b4187863759c1","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-35f8b93fae4b4187863759c1","content":"CLIP对称目标、候选集合、缓存、IVF及真实关系分数","sha256":"8b4a38fbe2d21e8c471d155a892cffc7decd7adfc064650236630fb3edf7a7d0","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-abc9aa30a017cb9017a93242","nextBlock":"b-2b0d51625875cdb99f08bd08","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2b0d51625875cdb99f08bd08","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-2b0d51625875cdb99f08bd08","content":"**学习路径：** [文字候选与图文检索](/course/study/clip-retrieval.html)","sha256":"bab4bd804571aa5afbaf197e9c1539510fc766578c9825aa46cadbb18444cc32","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-35f8b93fae4b4187863759c1","nextBlock":"b-37fbfe15a840fa86449d8fd0","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-37fbfe15a840fa86449d8fd0","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-37fbfe15a840fa86449d8fd0","content":"从[P57 配对矩阵](/course/pages/l3-r2-p57.html)进入，沿着[P58 文字候选](/course/pages/l3-r2-p58.html)到[P59 图库缓存](/course/pages/l3-r2-p59.html)：同一图文表示怎样从训练目标变成可用接口？","sha256":"50590b614b01aacf740b6f59f0a25be53dce523d2843825124cc6e1195f0f09c","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2b0d51625875cdb99f08bd08","nextBlock":"b-87a343db0d5cf1dac70e6445","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-87a343db0d5cf1dac70e6445","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-87a343db0d5cf1dac70e6445","content":"这里的zero-shot是**不使用当前下游数据集再训练模型，直接用已有图像和文字编码器比较候选描述**。不保证相关概念未进入预训练；没有专用分类头也不足以单独证明这个协议成立。","sha256":"2e7f9ca1ab75355dcb74d4a3f8cd95d8ec9fc0ddc39e8919ee0ca1d8fa2e2052","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-37fbfe15a840fa86449d8fd0","nextBlock":"b-53328cdd6d3c71b56eebfe62","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-53328cdd6d3c71b56eebfe62","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-53328cdd6d3c71b56eebfe62","content":"先看配对损失与候选分数，再到[LAB 04](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/0ba963931ddf4866b507b210d93a43bf)核对真实照片的cosine/softmax。关心检索时继续[LAB 12](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/921eeaf784284ce19568c6ca7619b55a)的缓存、命中ID和评价；近似索引及查询交互留第八讲选读。","sha256":"2b31d6a9ce476da9fbc54ff1cd84e4d23c0a209eef12ce6ab7054f0dfa6f5c76","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-87a343db0d5cf1dac70e6445","nextBlock":"b-ef2dd01ce398079aa5cbe781","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ef2dd01ce398079aa5cbe781","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-ef2dd01ce398079aa5cbe781","content":"读完应能区分表示质量、候选定义和索引误差：一个返回结果错了，是模型没排对，还是近似检索漏掉了本来排前的项？","sha256":"bfed64de92e5326d9f2ea08a5fbf216a9101f67d42e71710906c6ab6d2692320","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-53328cdd6d3c71b56eebfe62","nextBlock":"b-011878768e3079d2d392bcd5","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-011878768e3079d2d392bcd5","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-011878768e3079d2d392bcd5","content":"## 两个编码器，一个配对目标","sha256":"063457994938d4e5dfe61567141bd163cb7d7cce6959928c1c0eb5510469ab6b","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ef2dd01ce398079aa5cbe781","nextBlock":"b-5f730bbd8eff7ad501b1df09","prerequisites":[{"id":"b-2369c3f277c9225590898fa2","url":"#b-2369c3f277c9225590898fa2","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5f730bbd8eff7ad501b1df09","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-5f730bbd8eff7ad501b1df09","content":"图像与文字分别归一化为 $v_i,u_j$，矩阵 $S_{ij}=v_i^Tu_j$ 同时支持图搜文与文搜图。双向交叉熵让原有配对相对其他候选更容易被选出。课堂三对图文的完整数值见 [三对图文的双向损失：从单位向量到 0.09567](/course/blocks/clip-matrix-worked.html)：温度 .2，正确概率约 .90876，双向损失约 .09567。CLS03 热身采用另一组向量与温度，不能混用两套数值。","sha256":"421bb86160826d098ebd40ff162672a976ef2d422bd0fce72fa666c4eaad7d3e","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"v_i,u_j","definitionContext":"两个编码器，一个配对目标"},{"latex":"S_{ij}=v_i^Tu_j","definitionContext":"两个编码器，一个配对目标"}],"previousBlock":"b-011878768e3079d2d392bcd5","nextBlock":"b-3a497f200caba71b2da46c8e","prerequisites":[{"id":"b-011878768e3079d2d392bcd5","url":"#b-011878768e3079d2d392bcd5","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3a497f200caba71b2da46c8e","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-3a497f200caba71b2da46c8e","content":"## 为什么这是可用的新接口","sha256":"eba1b58a7fce19f59185999c38f5f654b59946a33ae131684e1abbc607f712e6","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5f730bbd8eff7ad501b1df09","nextBlock":"b-9833912d287f42801ed86c98","prerequisites":[{"id":"b-011878768e3079d2d392bcd5","url":"#b-011878768e3079d2d392bcd5","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9833912d287f42801ed86c98","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-9833912d287f42801ed86c98","content":"原始 CLIP 研究展示自然语言监督能迁移到多种视觉任务，并用文字定义当前分类候选；无需为每次候选集合重训一个固定类别头。[论文中的正面迁移证据](/course/blocks/clip-positive-transfer-evidence.html)。它留下的是图文比较接口，描述生成与定位仍需相应输出。","sha256":"6b5dab1b128812a45c0237b9943cd89a557634629a8386260132bba6eb7d7a14","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3a497f200caba71b2da46c8e","nextBlock":"b-4757d5a416e5d9d398b7ea86","prerequisites":[{"id":"b-3a497f200caba71b2da46c8e","url":"#b-3a497f200caba71b2da46c8e","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4757d5a416e5d9d398b7ea86","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-4757d5a416e5d9d398b7ea86","content":"## 三个干预，三个不同问题","sha256":"b82eb99f0b75c3c6d07cae11fdb8041ed3871643065e2b41ebf2f1d956f99de4","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9833912d287f42801ed86c98","nextBlock":"b-20adc2b98c5178f1be41aa69","prerequisites":[{"id":"b-3a497f200caba71b2da46c8e","url":"#b-3a497f200caba71b2da46c8e","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-20adc2b98c5178f1be41aa69","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-20adc2b98c5178f1be41aa69","content":"固定图像和候选文字，只新增选项：原余弦可保持，但 softmax 分母改变。改模板：文本向量改变，余弦也可能改变。改图片：视觉向量改变，应检查分数是否跟随所需视觉内容。","sha256":"eeb43459eb9f075a95c7403ff26d52e85c14cf632e9eef7d81193d82eb7eae86","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4757d5a416e5d9d398b7ea86","nextBlock":"b-c857dae38be3a3c7e43ae650","prerequisites":[{"id":"b-4757d5a416e5d9d398b7ea86","url":"#b-4757d5a416e5d9d398b7ea86","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c857dae38be3a3c7e43ae650","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-c857dae38be3a3c7e43ae650","content":"若猫/狗 logits 为 2/1，猫概率为 .7311；添加 logit 为2的“幼猫”，原猫项变为 .4223。两次图像与原猫文字并没有变。候选内概率不能单独解读为开放世界真实性。[完整算例](/course/blocks/clip-candidate-probability.html)。","sha256":"e2fd6ed94ff5812ee027b0c1d2a3d46127eb13403d0e3f6c267ac4ef37f49552","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-20adc2b98c5178f1be41aa69","nextBlock":"b-70c737fe2d1e5f5c4a362267","prerequisites":[{"id":"b-4757d5a416e5d9d398b7ea86","url":"#b-4757d5a416e5d9d398b7ea86","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-70c737fe2d1e5f5c4a362267","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3.html#b-70c737fe2d1e5f5c4a362267","content":"LAB04 的真实猫图已保存上述接口对照。LAB12 使用完整 512 维编码，不是把候选分数拼成向量。两本共享模型身份、各答不同问题：前者解释模型与候选，后者解释缓存和搜索。","sha256":"38c3c7217ff456579df20fe9c5703f47cfc3e63d9dcbb6bd611964eef7aa9ad0","chapter":{"id":"course","title":"两个编码器，一个配对目标"},"unit":{"id":"notebook-blocks/ed31be25c84b4c4eb511e2b846038ac3","title":"两个编码器，一个配对目标"},"page":{"id":"reading-opening","title":"两个编码器，一个配对目标","number":1},"sectionTitle":"两个编码器，一个配对目标","currentQuestion":"两个编码器，一个配对目标","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c857dae38be3a3c7e43ae650","nextBlock":null,"prerequisites":[{"id":"b-4757d5a416e5d9d398b7ea86","url":"#b-4757d5a416e5d9d398b7ea86","role":"section-title"},{"id":"b-c8d9497113fedff65cab6359","url":"#b-c8d9497113fedff65cab6359","role":"problem-context"}],"relatedBlocks":[]}]}
