{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"目标可以扩展，训练还要做得出来","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.md","pages":[{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","unit":"训练留下什么，训练答案又从哪里来？","unitId":"i-reuse-and-targets","chapter":"I","number":1}],"aliases":[{"oldId":"explanation","targetId":"h-i-objective-needs-computation","operation":"split-explanation-range"},{"oldId":"b-2e938686b4e3f26334d5021f","targetId":"b-2e938686b4e3f26334d5021f","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-6ffd8297f88398de63699f8d","targetId":"b-6ffd8297f88398de63699f8d","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"5afeda50c5711c277b07abe46d77aa665a73841f21f16a3c3033f8b428ca1cdb","afterSHA256":"5afeda50c5711c277b07abe46d77aa665a73841f21f16a3c3033f8b428ca1cdb","reason":"Same source body in authored semantic page"},{"oldId":"b-916f685f77bd4e5090a07032","targetId":"b-916f685f77bd4e5090a07032","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"a4c7eab21c83eccbcaefe98f60454f108c008c8b3fa6594af5b233ae84b8b0f1","afterSHA256":"a4c7eab21c83eccbcaefe98f60454f108c008c8b3fa6594af5b233ae84b8b0f1","reason":"Same source body in authored semantic page"},{"oldId":"b-7d84407f49033533e682b13f","targetId":"b-7d84407f49033533e682b13f","page":"h-i-objective-needs-computation","operation":"rewrite","beforeSHA256":"bb421f54480250ca2719e24391d2b418ce48c4d2a42bd6201f9e38d0f506a37b","afterSHA256":"b58fb684ccba3bf0b72e76aa8cc507b6a2535ba31c29d7d8130a3e9fa35de35a","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-c6b5396920eb76c7419a0f70","targetId":"b-c6b5396920eb76c7419a0f70","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-962b3700da03a592e2f336d9","targetId":"b-962b3700da03a592e2f336d9","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"1a451d4da39e0e8ea8b6b0bd27d302c64505173005e5df18856fe816695bdd1a","afterSHA256":"1a451d4da39e0e8ea8b6b0bd27d302c64505173005e5df18856fe816695bdd1a","reason":"Same source body in authored semantic page"},{"oldId":"b-e212c9f84dd7d589c670a465","targetId":"b-e212c9f84dd7d589c670a465","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"182b04104eb544a1d72fe491bb9fb983109e5743e186813a3f958018f77990d3","afterSHA256":"182b04104eb544a1d72fe491bb9fb983109e5743e186813a3f958018f77990d3","reason":"Same source body in authored semantic page"},{"oldId":"b-aada5fb94e9c56931450c2bf","targetId":"b-aada5fb94e9c56931450c2bf","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"41c0c8827ed3d1b42ab13467510bf5dd1ba108583478771eeb2f9851358a6055","afterSHA256":"41c0c8827ed3d1b42ab13467510bf5dd1ba108583478771eeb2f9851358a6055","reason":"Same source body in authored semantic page"},{"oldId":"b-85126862ec7f6cb826607b3d","targetId":"b-85126862ec7f6cb826607b3d","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"aa3396a1af2f8a274d0b98c3d282a23fe47342ee9bc0456df04dd8d50608f65b","afterSHA256":"aa3396a1af2f8a274d0b98c3d282a23fe47342ee9bc0456df04dd8d50608f65b","reason":"Same source body in authored semantic page"},{"oldId":"b-a9f8ba09ad801a9f412daf2f","targetId":"b-a9f8ba09ad801a9f412daf2f","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"ee7ba45d4a8a88485f4eb59743d22f646df82fe3ff7d2ac55977d48ae2b9ee94","afterSHA256":"ee7ba45d4a8a88485f4eb59743d22f646df82fe3ff7d2ac55977d48ae2b9ee94","reason":"Same source body in authored semantic page"},{"oldId":"b-3a7165f635e8a766e46acfbd","targetId":"b-3a7165f635e8a766e46acfbd","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-f0c33de74b9b10f92a92025e","targetId":"b-f0c33de74b9b10f92a92025e","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"af83bc941f07e4b1efefc49622821e537e57dbcb931dd219e3df734bf1afd0be","afterSHA256":"af83bc941f07e4b1efefc49622821e537e57dbcb931dd219e3df734bf1afd0be","reason":"Same source body in authored semantic page"},{"oldId":"b-e8f015ed450f1f9862211e77","targetId":"b-e8f015ed450f1f9862211e77","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"4416c8f8e74cb99082366ffa0ae6914277a75b017bf3d3a013dfb40c4ec6e379","afterSHA256":"4416c8f8e74cb99082366ffa0ae6914277a75b017bf3d3a013dfb40c4ec6e379","reason":"Same source body in authored semantic page"},{"oldId":"b-7005e43726458eb0e3ec4185","targetId":"b-7005e43726458eb0e3ec4185","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"0c7f02db3d18d117a5a4f235deddb2f35dc552c2d739ac75e21746e57768530b","afterSHA256":"0c7f02db3d18d117a5a4f235deddb2f35dc552c2d739ac75e21746e57768530b","reason":"Same source body in authored semantic page"},{"oldId":"b-6b876c4633dc762bc6f1a58f","targetId":"b-6b876c4633dc762bc6f1a58f","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"6ae953513cfb1f55bac140e74edb776cc264d1baa9263cd10c17724cb5b1d439","afterSHA256":"6ae953513cfb1f55bac140e74edb776cc264d1baa9263cd10c17724cb5b1d439","reason":"Same source body in authored semantic page"},{"oldId":"b-03ade65f6d39187e15cad2dc","targetId":"b-03ade65f6d39187e15cad2dc","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"523b99297d44fab7411d3c342d6fe4b25e9990ef616da8198b16bce20f9fea40","afterSHA256":"523b99297d44fab7411d3c342d6fe4b25e9990ef616da8198b16bce20f9fea40","reason":"Same source body in authored semantic page"},{"oldId":"b-2a9497579f55db97ceea9a87","targetId":"b-2a9497579f55db97ceea9a87","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"745e6edf27683b6f3da53729a1368786760ded1ad08af544b11c2932e1890ac2","afterSHA256":"745e6edf27683b6f3da53729a1368786760ded1ad08af544b11c2932e1890ac2","reason":"Same source body in authored semantic page"},{"oldId":"b-f402f17b734c7103e8c2019c","targetId":"b-f402f17b734c7103e8c2019c","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"38301981e800d30ae5ffdc779a46cfbd0e9ca74c7b85b55c409ebd875d709a1b","afterSHA256":"38301981e800d30ae5ffdc779a46cfbd0e9ca74c7b85b55c409ebd875d709a1b","reason":"Same source body in authored semantic page"},{"oldId":"b-bb358977ced3e4799b269e43","targetId":"b-bb358977ced3e4799b269e43","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"577dd13d9bdb28c29f839cceac0acce539f96fbadf2e9a8037fb1a67cd871c51","afterSHA256":"577dd13d9bdb28c29f839cceac0acce539f96fbadf2e9a8037fb1a67cd871c51","reason":"Same source body in authored semantic page"},{"oldId":"b-a04c3ce38ecaae42bd082faa","targetId":"b-a04c3ce38ecaae42bd082faa","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"0571dca9acd087bd8dd6348c46ea4bca912c74f04739403b799ab79950538ed9","afterSHA256":"0571dca9acd087bd8dd6348c46ea4bca912c74f04739403b799ab79950538ed9","reason":"Same source body in authored semantic page"},{"oldId":"b-e92b7b9205e402261f26ff1a","targetId":"b-e92b7b9205e402261f26ff1a","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"0303b1b76d7ba697d417fd6f96051a0609a330d4b1fb646413546de9845b2f12","afterSHA256":"0303b1b76d7ba697d417fd6f96051a0609a330d4b1fb646413546de9845b2f12","reason":"Same source body in authored semantic page"},{"oldId":"b-e9c00fad11b75c939fba0f1f","targetId":"b-e9c00fad11b75c939fba0f1f","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"082a76b80dda3ebbe515fd869a91d9a862bf7abd7c086e286f3511b424009378","afterSHA256":"082a76b80dda3ebbe515fd869a91d9a862bf7abd7c086e286f3511b424009378","reason":"Same source body in authored semantic page"},{"oldId":"b-c8e26ff6dc7d4249c4cf6ca8","targetId":"b-c8e26ff6dc7d4249c4cf6ca8","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"39d25dd4fe2f52149cee1f6dd66c2824cbab82e5abc1c4c93522b42ad8cde2ce","afterSHA256":"39d25dd4fe2f52149cee1f6dd66c2824cbab82e5abc1c4c93522b42ad8cde2ce","reason":"Same source body in authored semantic page"},{"oldId":"b-de7d9febaec141a997cd16dd","targetId":"b-de7d9febaec141a997cd16dd","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"38b9d3070b7c308d91bf2eba039f8adc43a3f983cccb65f69453d4b5070f18de","afterSHA256":"38b9d3070b7c308d91bf2eba039f8adc43a3f983cccb65f69453d4b5070f18de","reason":"Same source body in authored semantic page"},{"oldId":"b-631a50db87a83d4a197d8199","targetId":"b-631a50db87a83d4a197d8199","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"eb4d86f9172101d0dce0336f8eb3224dcb037c940d0f939c2a796864984244b0","afterSHA256":"eb4d86f9172101d0dce0336f8eb3224dcb037c940d0f939c2a796864984244b0","reason":"Same source body in authored semantic page"},{"oldId":"b-805a3c143093a543097f63dc","targetId":"b-805a3c143093a543097f63dc","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"ecad0fc50b77cc07b03d9b41a5d58ab38bfd17b3128d8d7814f9222eff1e6d08","afterSHA256":"ecad0fc50b77cc07b03d9b41a5d58ab38bfd17b3128d8d7814f9222eff1e6d08","reason":"Same source body in authored semantic page"},{"oldId":"b-0f2b50a200c7317c4d27b309","targetId":"b-0f2b50a200c7317c4d27b309","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"d5e28e5cc461a31aaf3839cdaa1ef550910922f88e4a1251d27c33a1e35ab591","afterSHA256":"d5e28e5cc461a31aaf3839cdaa1ef550910922f88e4a1251d27c33a1e35ab591","reason":"Same source body in authored semantic page"},{"oldId":"b-a0b1d0d90c0c79f42a814fe7","targetId":"b-a0b1d0d90c0c79f42a814fe7","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"73df72ed0fa244b2e7c8840ba245f363dec994a0bde13e7064edc4ff5888ebf5","afterSHA256":"73df72ed0fa244b2e7c8840ba245f363dec994a0bde13e7064edc4ff5888ebf5","reason":"Same source body in authored semantic page"},{"oldId":"b-25934d73c1402a664cf8ffe8","targetId":"b-25934d73c1402a664cf8ffe8","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"7440ea19dd094aacb24dbaf490eb5ef3980388ed936a9a6ab7b9dafc7330cc54","afterSHA256":"7440ea19dd094aacb24dbaf490eb5ef3980388ed936a9a6ab7b9dafc7330cc54","reason":"Same source body in authored semantic page"},{"oldId":"b-c9ed0efef4c2ac65c95d6b78","targetId":"b-c9ed0efef4c2ac65c95d6b78","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"ac69c70eea43a03f90404143576e3704e1f25c2808567a85e9dade6f8819e028","afterSHA256":"ac69c70eea43a03f90404143576e3704e1f25c2808567a85e9dade6f8819e028","reason":"Same source body in authored semantic page"},{"oldId":"b-4a155781845b62637d1590ae","targetId":"b-4a155781845b62637d1590ae","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"737a7630b679edb7747534f321a93cddbbc752208f9d32a8c0a6df81b0f1e4df","afterSHA256":"737a7630b679edb7747534f321a93cddbbc752208f9d32a8c0a6df81b0f1e4df","reason":"Same source body in authored semantic page"},{"oldId":"b-3d2013cd6a0637bcf85c4efc","targetId":"b-3d2013cd6a0637bcf85c4efc","page":"h-i-objective-needs-computation","operation":"move","beforeSHA256":"4f595d6fbd9089e01a325869faff81aec44c6a34d593ce8788ad05a6d2754eba","afterSHA256":"4f595d6fbd9089e01a325869faff81aec44c6a34d593ce8788ad05a6d2754eba","reason":"Same source body in authored semantic page"},{"oldId":"h-i-objective-needs-computation-title","targetId":"b-54f40ad2bbae514c55b730a2","operation":"retain-source-identity"}],"blocks":[{"id":"b-54f40ad2bbae514c55b730a2","type":"h2","scope":"h-i-objective-needs-computation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-54f40ad2bbae514c55b730a2","content":"## 训练目标还要落到一次可执行的更新","sha256":"765de492e6a6f943847749154aa86712cb69c5ea60ea742265eb8ee60a47c805","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-2e938686b4e3f26334d5021f","prerequisites":[{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-2e938686b4e3f26334d5021f","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-2e938686b4e3f26334d5021f","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-54f40ad2bbae514c55b730a2","nextBlock":"b-6ffd8297f88398de63699f8d","prerequisites":[{"id":"b-54f40ad2bbae514c55b730a2","url":"#b-54f40ad2bbae514c55b730a2","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-6ffd8297f88398de63699f8d","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-6ffd8297f88398de63699f8d","content":"# P07 · 目标可以扩展，训练还要做得出来","sha256":"5afeda50c5711c277b07abe46d77aa665a73841f21f16a3c3033f8b428ca1cdb","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2e938686b4e3f26334d5021f","nextBlock":"b-916f685f77bd4e5090a07032","prerequisites":[{"id":"b-2e938686b4e3f26334d5021f","url":"#b-2e938686b4e3f26334d5021f","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-916f685f77bd4e5090a07032","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-916f685f77bd4e5090a07032","content":"资源约束在本讲是方法选择的一个条件；真正需要回访它的是批内候选、MoCo缓存与MAE可见块。","sha256":"a4c7eab21c83eccbcaefe98f60454f108c008c8b3fa6594af5b233ae84b8b0f1","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6ffd8297f88398de63699f8d","nextBlock":"b-7d84407f49033533e682b13f","prerequisites":[{"id":"b-6ffd8297f88398de63699f8d","url":"#b-6ffd8297f88398de63699f8d","role":"section-title"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-7d84407f49033533e682b13f","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-7d84407f49033533e682b13f","content":"图内标签与数值：P07 学习表示机制图，状态 1；学习目标；训练什么；数据供给；拿到什么；计算条件；算得起什么；一起决定可行的实验；当前批次：更多图片；历史字典：复用特征；可见图块：减少编码\n\n目标、数据、计算三个相连节点，以及当前批次、历史特征、可见图块三个具体计算取舍。\n\n- 学习目标\n- 训练什么\n- 数据供给\n- 拿到什么\n- 计算条件\n- 算得起什么\n- 一起决定可行的实验\n- 当前批次：更多图片\n- 历史字典：复用特征\n- 可见图块：减少编码目标、数据、计算三个相连节点，以及当前批次、历史特征、可见图块三个具体计算取舍。 [查看对应图解状态](/#/l3-r2-p07/0)[查看完整图解](/course/figures/l3-r2-p07-0.html)","sha256":"b58fb684ccba3bf0b72e76aa8cc507b6a2535ba31c29d7d8130a3e9fa35de35a","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"目标、数据、计算三个相连节点，以及当前批次、历史特征、可见图块三个具体计算取舍。 查看对应图解状态","labels":["学习目标","训练什么","数据供给","拿到什么","计算条件","算得起什么","一起决定可行的实验","当前批次：更多图片","历史字典：复用特征","可见图块：减少编码"],"description":"目标、数据、计算三个相连节点，以及当前批次、历史特征、可见图块三个具体计算取舍。学习目标训练什么数据供给拿到什么计算条件算得起什么一起决定可行的实验当前批次：更多图片历史字典：复用特征可见图块：减少编码","evidenceIdentity":"目标、数据、计算三个相连节点，以及当前批次、历史特征、可见图块三个具体计算取舍。 查看对应图解状态","images":[]},"previousBlock":"b-916f685f77bd4e5090a07032","nextBlock":"b-c6b5396920eb76c7419a0f70","prerequisites":[{"id":"b-6ffd8297f88398de63699f8d","url":"#b-6ffd8297f88398de63699f8d","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c6b5396920eb76c7419a0f70","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-c6b5396920eb76c7419a0f70","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7d84407f49033533e682b13f","nextBlock":"b-962b3700da03a592e2f336d9","prerequisites":[{"id":"b-6ffd8297f88398de63699f8d","url":"#b-6ffd8297f88398de63699f8d","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-962b3700da03a592e2f336d9","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-962b3700da03a592e2f336d9","content":"找到可扩展的目标以后，训练也不会变成免费的。更多数据、更大的网络、更长的训练，仍然需要存储、计算和时间；设备之间合作，也有额外代价。\n但这不是一页“模型很烧钱”的背景介绍。资源条件会影响方法怎么设计：候选特征能不能从历史批次复用？被遮住的图块是不是也要送进昂贵的编码器？这些选择会改变一次实验能够使用多少材料。\n我们先记住这几个具体问题，后面看 MoCo 和 MAE 时就能回到它们。至于扩展定律、多设备分片和通信细节，属于后续系统展开的内容。今天只需要理解：好的学习目标，还需要一种可实现的计算组织。","sha256":"1a451d4da39e0e8ea8b6b0bd27d302c64505173005e5df18856fe816695bdd1a","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c6b5396920eb76c7419a0f70","nextBlock":"b-e212c9f84dd7d589c670a465","prerequisites":[{"id":"b-c6b5396920eb76c7419a0f70","url":"#b-c6b5396920eb76c7419a0f70","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-e212c9f84dd7d589c670a465","type":"h3","scope":"objective-and-compute-interface","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-e212c9f84dd7d589c670a465","content":"### 训练目标也在分配计算","sha256":"182b04104eb544a1d72fe491bb9fb983109e5743e186813a3f958018f77990d3","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标也在分配计算","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":"objective-and-compute-interface","formulas":[],"previousBlock":"b-962b3700da03a592e2f336d9","nextBlock":"b-aada5fb94e9c56931450c2bf","prerequisites":[{"id":"b-c6b5396920eb76c7419a0f70","url":"#b-c6b5396920eb76c7419a0f70","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"},{"id":"b-aada5fb94e9c56931450c2bf","url":"#b-aada5fb94e9c56931450c2bf","relation":"same-reasoning"},{"id":"b-85126862ec7f6cb826607b3d","url":"#b-85126862ec7f6cb826607b3d","relation":"same-reasoning"},{"id":"b-a9f8ba09ad801a9f412daf2f","url":"#b-a9f8ba09ad801a9f412daf2f","relation":"same-reasoning"}]},{"id":"b-aada5fb94e9c56931450c2bf","type":"p","scope":"objective-and-compute-interface","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-aada5fb94e9c56931450c2bf","content":"扩大目标供给后，还要把训练真正做出来。当前批内对比会同时保留多张图两种视图的前向和反向状态；保存历史key的MoCo用特征缓存减少重复编码，但必须处理旧特征的一致性；MAE让较重编码器只处理可见图块，把补全位置的工作放到较轻解码器。","sha256":"41c0c8827ed3d1b42ab13467510bf5dd1ba108583478771eeb2f9851358a6055","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标也在分配计算","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":"objective-and-compute-interface","formulas":[],"previousBlock":"b-e212c9f84dd7d589c670a465","nextBlock":"b-85126862ec7f6cb826607b3d","prerequisites":[{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"},{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","relation":"same-reasoning"},{"id":"b-85126862ec7f6cb826607b3d","url":"#b-85126862ec7f6cb826607b3d","relation":"same-reasoning"},{"id":"b-a9f8ba09ad801a9f412daf2f","url":"#b-a9f8ba09ad801a9f412daf2f","relation":"same-reasoning"}]},{"id":"b-85126862ec7f6cb826607b3d","type":"p","scope":"objective-and-compute-interface","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-85126862ec7f6cb826607b3d","content":"所以计算条件会参与方法设计，而不只是最后购买多少设备。我们可以先画资源账：本步编码多少新图，保留多少激活，缓存多少旧向量，下步需要重复什么。这个账足以理解接下来的batch、队列与可见块取舍。","sha256":"aa3396a1af2f8a274d0b98c3d282a23fe47342ee9bc0456df04dd8d50608f65b","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标也在分配计算","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":"objective-and-compute-interface","formulas":[],"previousBlock":"b-aada5fb94e9c56931450c2bf","nextBlock":"b-a9f8ba09ad801a9f412daf2f","prerequisites":[{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"},{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","relation":"same-reasoning"},{"id":"b-aada5fb94e9c56931450c2bf","url":"#b-aada5fb94e9c56931450c2bf","relation":"same-reasoning"},{"id":"b-a9f8ba09ad801a9f412daf2f","url":"#b-a9f8ba09ad801a9f412daf2f","relation":"same-reasoning"}]},{"id":"b-a9f8ba09ad801a9f412daf2f","type":"p","scope":"objective-and-compute-interface","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-a9f8ba09ad801a9f412daf2f","content":"本讲不在此展开并行算法、显存公式或分片通信。它们保留为第六讲预备阅读；这里带走的是同一个原则：**需要更多比较对象，不一定等于同时反传更多图片；需要预测更多位置，也不一定等于让昂贵编码器看见所有位置。**","sha256":"ee7ba45d4a8a88485f4eb59743d22f646df82fe3ff7d2ac55977d48ae2b9ee94","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标也在分配计算","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":"objective-and-compute-interface","formulas":[],"previousBlock":"b-85126862ec7f6cb826607b3d","nextBlock":"b-3a7165f635e8a766e46acfbd","prerequisites":[{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"},{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","relation":"same-reasoning"},{"id":"b-aada5fb94e9c56931450c2bf","url":"#b-aada5fb94e9c56931450c2bf","relation":"same-reasoning"},{"id":"b-85126862ec7f6cb826607b3d","url":"#b-85126862ec7f6cb826607b3d","relation":"same-reasoning"}]},{"id":"b-3a7165f635e8a766e46acfbd","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-3a7165f635e8a766e46acfbd","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a9f8ba09ad801a9f412daf2f","nextBlock":"b-f0c33de74b9b10f92a92025e","prerequisites":[{"id":"b-e212c9f84dd7d589c670a465","url":"#b-e212c9f84dd7d589c670a465","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-f0c33de74b9b10f92a92025e","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-f0c33de74b9b10f92a92025e","content":"先读本页计算接口，直接进入P8；多设备与分片算例是L06预备选读。","sha256":"af83bc941f07e4b1efefc49622821e537e57dbcb931dd219e3df734bf1afd0be","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3a7165f635e8a766e46acfbd","nextBlock":"b-f5a215a61628d9130f8dc948","prerequisites":[{"id":"b-3a7165f635e8a766e46acfbd","url":"#b-3a7165f635e8a766e46acfbd","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-f5a215a61628d9130f8dc948","type":"h4","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-f5a215a61628d9130f8dc948","content":"#### 按问题回查的讲解与短练习（1）","sha256":"0480869f0a66009b015a757d67aed6061011e378987f98a024f4e8bfc5386152","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f0c33de74b9b10f92a92025e","nextBlock":"b-8a84741887b074ad620c1521","prerequisites":[{"id":"b-3a7165f635e8a766e46acfbd","url":"#b-3a7165f635e8a766e46acfbd","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-8a84741887b074ad620c1521","type":"h4","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-8a84741887b074ad620c1521","content":"#### 这条线索的展开与后续阅读","sha256":"48b8229f7559b36ed4d5628416e3991e5c73293fbce43374db67386c8736f3bc","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f5a215a61628d9130f8dc948","nextBlock":"b-e8f015ed450f1f9862211e77","prerequisites":[{"id":"b-f5a215a61628d9130f8dc948","url":"#b-f5a215a61628d9130f8dc948","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-e8f015ed450f1f9862211e77","type":"ul","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-e8f015ed450f1f9862211e77","content":"- [参数、数据与训练量，要一起分配](/supplement.html?return=l3-r2-p07/0#/l3-compute-budget/0) 第六讲相关预备阅读\n- [两张卡处理不同数据，更新同一组参数](/supplement.html?return=l3-r2-p07/0#/l3-data-parallel/0) 第六讲相关预备阅读\n- [放不下时，把常驻训练状态分摊开](/supplement.html?return=l3-r2-p07/0#/l3-sharding/0) 第六讲相关预备阅读\n- [学习目标，也在决定计算花在哪里](/supplement.html?return=l3-r2-p07/0#/l3-compute-compare/0) L03补充阅读\n- [多个数据分片仍是一回更新｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-problem/0) 第六讲相关预备阅读\n- [多个数据分片仍是一回更新｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-mechanism/0) 第六讲相关预备阅读\n- [多个数据分片仍是一回更新｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-example/0) 第六讲相关预备阅读\n- [多个数据分片仍是一回更新｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-evidence/0) 第六讲相关预备阅读\n- [多个数据分片仍是一回更新｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-boundary/0) 第六讲相关预备阅读\n- [显存里不只是一份权重｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-problem/0) 第六讲相关预备阅读\n- [显存里不只是一份权重｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-mechanism/0) 第六讲相关预备阅读\n- [显存里不只是一份权重｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-example/0) 第六讲相关预备阅读\n- [显存里不只是一份权重｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-evidence/0) 第六讲相关预备阅读\n- [显存里不只是一份权重｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-boundary/0) 第六讲相关预备阅读\n- [分片省状态，也制造通信｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-problem/0) 第六讲相关预备阅读\n- [分片省状态，也制造通信｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-mechanism/0) 第六讲相关预备阅读\n- [分片省状态，也制造通信｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-example/0) 第六讲相关预备阅读\n- [分片省状态，也制造通信｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-evidence/0) 第六讲相关预备阅读\n- [分片省状态，也制造通信｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-boundary/0) 第六讲相关预备阅读","sha256":"4416c8f8e74cb99082366ffa0ae6914277a75b017bf3d3a013dfb40c4ec6e379","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8a84741887b074ad620c1521","nextBlock":"b-7005e43726458eb0e3ec4185","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-7005e43726458eb0e3ec4185","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-7005e43726458eb0e3ec4185","content":"- [参数、数据与训练量，要一起分配](/supplement.html?return=l3-r2-p07/0#/l3-compute-budget/0) 第六讲相关预备阅读","sha256":"0c7f02db3d18d117a5a4f235deddb2f35dc552c2d739ac75e21746e57768530b","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e8f015ed450f1f9862211e77","nextBlock":"b-6b876c4633dc762bc6f1a58f","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-6b876c4633dc762bc6f1a58f","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-6b876c4633dc762bc6f1a58f","content":"- [两张卡处理不同数据，更新同一组参数](/supplement.html?return=l3-r2-p07/0#/l3-data-parallel/0) 第六讲相关预备阅读","sha256":"6ae953513cfb1f55bac140e74edb776cc264d1baa9263cd10c17724cb5b1d439","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7005e43726458eb0e3ec4185","nextBlock":"b-03ade65f6d39187e15cad2dc","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-03ade65f6d39187e15cad2dc","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-03ade65f6d39187e15cad2dc","content":"- [放不下时，把常驻训练状态分摊开](/supplement.html?return=l3-r2-p07/0#/l3-sharding/0) 第六讲相关预备阅读","sha256":"523b99297d44fab7411d3c342d6fe4b25e9990ef616da8198b16bce20f9fea40","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6b876c4633dc762bc6f1a58f","nextBlock":"b-2a9497579f55db97ceea9a87","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-2a9497579f55db97ceea9a87","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-2a9497579f55db97ceea9a87","content":"- [学习目标，也在决定计算花在哪里](/supplement.html?return=l3-r2-p07/0#/l3-compute-compare/0) L03补充阅读","sha256":"745e6edf27683b6f3da53729a1368786760ded1ad08af544b11c2932e1890ac2","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-03ade65f6d39187e15cad2dc","nextBlock":"b-f402f17b734c7103e8c2019c","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-f402f17b734c7103e8c2019c","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-f402f17b734c7103e8c2019c","content":"- [多个数据分片仍是一回更新｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-problem/0) 第六讲相关预备阅读","sha256":"38301981e800d30ae5ffdc779a46cfbd0e9ca74c7b85b55c409ebd875d709a1b","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2a9497579f55db97ceea9a87","nextBlock":"b-bb358977ced3e4799b269e43","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-bb358977ced3e4799b269e43","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-bb358977ced3e4799b269e43","content":"- [多个数据分片仍是一回更新｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-mechanism/0) 第六讲相关预备阅读","sha256":"577dd13d9bdb28c29f839cceac0acce539f96fbadf2e9a8037fb1a67cd871c51","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f402f17b734c7103e8c2019c","nextBlock":"b-a04c3ce38ecaae42bd082faa","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-a04c3ce38ecaae42bd082faa","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-a04c3ce38ecaae42bd082faa","content":"- [多个数据分片仍是一回更新｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-example/0) 第六讲相关预备阅读","sha256":"0571dca9acd087bd8dd6348c46ea4bca912c74f04739403b799ab79950538ed9","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-bb358977ced3e4799b269e43","nextBlock":"b-e92b7b9205e402261f26ff1a","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-e92b7b9205e402261f26ff1a","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-e92b7b9205e402261f26ff1a","content":"- [多个数据分片仍是一回更新｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-evidence/0) 第六讲相关预备阅读","sha256":"0303b1b76d7ba697d417fd6f96051a0609a330d4b1fb646413546de9845b2f12","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a04c3ce38ecaae42bd082faa","nextBlock":"b-e9c00fad11b75c939fba0f1f","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-e9c00fad11b75c939fba0f1f","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-e9c00fad11b75c939fba0f1f","content":"- [多个数据分片仍是一回更新｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-distributed-update-boundary/0) 第六讲相关预备阅读","sha256":"082a76b80dda3ebbe515fd869a91d9a862bf7abd7c086e286f3511b424009378","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e92b7b9205e402261f26ff1a","nextBlock":"b-c8e26ff6dc7d4249c4cf6ca8","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-c8e26ff6dc7d4249c4cf6ca8","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-c8e26ff6dc7d4249c4cf6ca8","content":"- [显存里不只是一份权重｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-problem/0) 第六讲相关预备阅读","sha256":"39d25dd4fe2f52149cee1f6dd66c2824cbab82e5abc1c4c93522b42ad8cde2ce","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e9c00fad11b75c939fba0f1f","nextBlock":"b-de7d9febaec141a997cd16dd","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-de7d9febaec141a997cd16dd","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-de7d9febaec141a997cd16dd","content":"- [显存里不只是一份权重｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-mechanism/0) 第六讲相关预备阅读","sha256":"38b9d3070b7c308d91bf2eba039f8adc43a3f983cccb65f69453d4b5070f18de","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c8e26ff6dc7d4249c4cf6ca8","nextBlock":"b-631a50db87a83d4a197d8199","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-631a50db87a83d4a197d8199","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-631a50db87a83d4a197d8199","content":"- [显存里不只是一份权重｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-example/0) 第六讲相关预备阅读","sha256":"eb4d86f9172101d0dce0336f8eb3224dcb037c940d0f939c2a796864984244b0","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-de7d9febaec141a997cd16dd","nextBlock":"b-805a3c143093a543097f63dc","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-805a3c143093a543097f63dc","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-805a3c143093a543097f63dc","content":"- [显存里不只是一份权重｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-evidence/0) 第六讲相关预备阅读","sha256":"ecad0fc50b77cc07b03d9b41a5d58ab38bfd17b3128d8d7814f9222eff1e6d08","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-631a50db87a83d4a197d8199","nextBlock":"b-0f2b50a200c7317c4d27b309","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-0f2b50a200c7317c4d27b309","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-0f2b50a200c7317c4d27b309","content":"- [显存里不只是一份权重｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-training-memory-boundary/0) 第六讲相关预备阅读","sha256":"d5e28e5cc461a31aaf3839cdaa1ef550910922f88e4a1251d27c33a1e35ab591","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-805a3c143093a543097f63dc","nextBlock":"b-a0b1d0d90c0c79f42a814fe7","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-a0b1d0d90c0c79f42a814fe7","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-a0b1d0d90c0c79f42a814fe7","content":"- [分片省状态，也制造通信｜问题从哪里出现](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-problem/0) 第六讲相关预备阅读","sha256":"73df72ed0fa244b2e7c8840ba245f363dec994a0bde13e7064edc4ff5888ebf5","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0f2b50a200c7317c4d27b309","nextBlock":"b-25934d73c1402a664cf8ffe8","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-25934d73c1402a664cf8ffe8","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-25934d73c1402a664cf8ffe8","content":"- [分片省状态，也制造通信｜核心改动怎样工作](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-mechanism/0) 第六讲相关预备阅读","sha256":"7440ea19dd094aacb24dbaf490eb5ef3980388ed936a9a6ab7b9dafc7330cc54","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a0b1d0d90c0c79f42a814fe7","nextBlock":"b-c9ed0efef4c2ac65c95d6b78","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-c9ed0efef4c2ac65c95d6b78","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-c9ed0efef4c2ac65c95d6b78","content":"- [分片省状态，也制造通信｜一起算一个完整例子](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-example/0) 第六讲相关预备阅读","sha256":"ac69c70eea43a03f90404143576e3704e1f25c2808567a85e9dade6f8819e028","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-25934d73c1402a664cf8ffe8","nextBlock":"b-4a155781845b62637d1590ae","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-4a155781845b62637d1590ae","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-4a155781845b62637d1590ae","content":"- [分片省状态，也制造通信｜怎样检查这项解释](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-evidence/0) 第六讲相关预备阅读","sha256":"737a7630b679edb7747534f321a93cddbbc752208f9d32a8c0a6df81b0f1e4df","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c9ed0efef4c2ac65c95d6b78","nextBlock":"b-3d2013cd6a0637bcf85c4efc","prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]},{"id":"b-3d2013cd6a0637bcf85c4efc","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p07.html#b-3d2013cd6a0637bcf85c4efc","content":"- [分片省状态，也制造通信｜什么结论还不能推出](/supplement.html?return=l3-r2-p07/0#/l3-sharding-communication-boundary/0) 第六讲相关预备阅读","sha256":"4f595d6fbd9089e01a325869faff81aec44c6a34d593ce8788ad05a6d2754eba","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"h-i-objective-needs-computation","title":"训练目标还要落到一次可执行的更新","number":1},"sectionTitle":"训练目标还要落到一次可执行的更新","currentQuestion":"训练留下什么，训练答案又从哪里来？","sourceSlides":["l3-r2-p07"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4a155781845b62637d1590ae","nextBlock":null,"prerequisites":[{"id":"b-8a84741887b074ad620c1521","url":"#b-8a84741887b074ad620c1521","role":"section-title"},{"id":"b-916f685f77bd4e5090a07032","url":"#b-916f685f77bd4e5090a07032","role":"problem-context"}],"relatedBlocks":[{"id":"b-7d84407f49033533e682b13f","url":"#b-7d84407f49033533e682b13f","relation":"figure-and-explanation"}]}]}