{"id":4549,"date":"2026-05-06T21:00:32","date_gmt":"2026-05-06T13:00:32","guid":{"rendered":"https:\/\/1ow.top\/index.php\/2026\/05\/06\/%e6%99%ba%e8%b0%b1%e5%85%ac%e5%b8%83%e9%99%8d%e6%99%ba%e7%9a%84%e7%a7%98%e5%af%86%ef%bc%9ascaling%e4%b8%8d%e5%8f%af%e9%81%bf%e5%85%8d%e7%9a%84%e7%97%9b\/"},"modified":"2026-05-06T21:00:32","modified_gmt":"2026-05-06T13:00:32","slug":"%e6%99%ba%e8%b0%b1%e5%85%ac%e5%b8%83%e9%99%8d%e6%99%ba%e7%9a%84%e7%a7%98%e5%af%86%ef%bc%9ascaling%e4%b8%8d%e5%8f%af%e9%81%bf%e5%85%8d%e7%9a%84%e7%97%9b","status":"publish","type":"post","link":"https:\/\/1ow.top\/index.php\/2026\/05\/06\/%e6%99%ba%e8%b0%b1%e5%85%ac%e5%b8%83%e9%99%8d%e6%99%ba%e7%9a%84%e7%a7%98%e5%af%86%ef%bc%9ascaling%e4%b8%8d%e5%8f%af%e9%81%bf%e5%85%8d%e7%9a%84%e7%97%9b\/","title":{"rendered":"\u667a\u8c31\u516c\u5e03\u201c\u964d\u667a\u201d\u7684\u79d8\u5bc6\uff1aScaling\u4e0d\u53ef\u907f\u514d\u7684\u75db"},"content":{"rendered":"<figure class=\"wp-block-image\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/eb9601d51374115f831ffb00d0abd13b.webp'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  decoding=\"async\" data-original=\"https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/eb9601d51374115f831ffb00d0abd13b.webp\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\u667a\u8c31\u516c\u5e03\u201c\u964d\u667a\u201d\u7684\u79d8\u5bc6\uff1aScaling\u4e0d\u53ef\u907f\u514d\u7684\u75db\"\/><\/div><\/figure>\n<figure class=\"wp-block-image\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/18770943747808d12b2c74f65d372c2f.webp'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  decoding=\"async\" data-original=\"https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/18770943747808d12b2c74f65d372c2f.webp\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\u667a\u8c31\u516c\u5e03\u201c\u964d\u667a\u201d\u7684\u79d8\u5bc6\uff1aScaling\u4e0d\u53ef\u907f\u514d\u7684\u75db\"\/><\/div><\/figure>\n<figure class=\"wp-block-image\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/405535242a104f16074f3fff537e15b0.jpeg'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  decoding=\"async\" data-original=\"https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/405535242a104f16074f3fff537e15b0.jpeg\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\u667a\u8c31\u516c\u5e03\u201c\u964d\u667a\u201d\u7684\u79d8\u5bc6\uff1aScaling\u4e0d\u53ef\u907f\u514d\u7684\u75db\"\/><\/div><\/figure>\n<figure class=\"wp-block-image\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/549dd092a858b9f110d3c7aff522c838.webp'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  decoding=\"async\" data-original=\"https:\/\/1ow.top\/wp-content\/uploads\/2026\/05\/549dd092a858b9f110d3c7aff522c838.webp\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\u667a\u8c31\u516c\u5e03\u201c\u964d\u667a\u201d\u7684\u79d8\u5bc6\uff1aScaling\u4e0d\u53ef\u907f\u514d\u7684\u75db\"\/><\/div><\/figure>\n<p>\u9e6d\u7fbd \u53d1\u81ea \u51f9\u975e\u5bfa<\/p>\n<p>Scaling\u5373\u6b63\u4e49\uff1f\u667a\u8c31\u6320\u4e86\u6320\u5934\u2014\u2014<\/p>\n<p>\u5f88\u75db\u82e6\uff0c\u800c\u4e14\u538b\u529b\u5c71\u5927\u2026\u2026<\/p>\n<p>\u667a\u8c31\u6700\u65b0\u53d1\u5e03\u7684\u4e00\u7bc7\u6280\u672f\u535a\u5ba2\uff0c\u753b\u98ce\u7a0d\u5fae\u6709\u70b9\u4e0d\u4e00\u6837\uff1a<\/p>\n<p>\u6ca1\u6709\u8fc7\u53bb\u7684\u786c\u6838\u6280\u672f\u8f93\u51fa\uff0c\u53cd\u800c\u5927\u5012\u82e6\u6c34\u4eceGLM-5\u4ee5\u6765\u7684\u5404\u79cd\u82b1\u5f0f\u8e29\u5751\uff0c\u5b98\u65b9\u79f0\u4e4b\u4e3a\u300cScaling Pain\u300d\u3002<\/p>\n<p>\u6211\u4eec\u7684\u63a8\u7406\u57fa\u7840\u8bbe\u65bd\u6b63\u627f\u53d7\u7740\u524d\u6240\u672a\u6709\u7684\u538b\u529b\uff0c\u6bcf\u5929\u90fd\u8981\u670d\u52a1\u6570\u4ebf\u6b21Coding Agent\u8c03\u7528\u3002<\/p>\n<p>\u8fc7\u53bb\u51e0\u5468\uff0c\u4e00\u4e9b\u7528\u6237\u5728\u4f7f\u7528GLM-5\u7cfb\u5217\u6a21\u578b\u6267\u884c\u590d\u6742Coding Agent\u4efb\u52a1\u65f6\uff0c\u906d\u9047\u591a\u79cd\u5f02\u5e38\uff0c\u6bd4\u5982\u4e71\u7801\u3001\u590d\u8bfb\u548c\u7f55\u89c1\u5b57\u7b26\u751f\u6210\u3002<\/p>\n<p>\u800c\u4e14\u8fd9\u4e9b\u95ee\u9898\u5728\u6807\u51c6\u63a8\u7406\u73af\u5883\u4e2d\u538b\u6839\u590d\u73b0\u4e0d\u51fa\u6765\uff01\uff01\uff01<\/p>\n<p>\u6392\u67e5\u6570\u5468\uff0c\u56e2\u961f\u7ec8\u4e8e\u63ea\u51fa\u771f\u51f6\uff0c\u5f7b\u5e95\u6233\u7834Scaling Laws\u8def\u4e0a\u7684\u9690\u5f62Bug\u3002<\/p>\n<p>\u4e0d\u4ec5\u8be6\u7ec6\u603b\u7ed3\u4e86\u81ea\u8eab\u906d\u9047\u7684\u6602\u8d35\u6559\u8bad\uff0c\u8fd8\u7ed9\u51fa\u4e86\u4e00\u5957\u6781\u5177\u5b9e\u64cd\u6027\u7684\u907f\u5751\u6307\u5357\u3002<\/p>\n<p>\u7b80\u5355\u6765\u8bf4\uff0c\u5982\u679c\u5c4f\u5e55\u524d\u7684\u4f60\u6b63\u6253\u7b97\u7ed9\u81ea\u5df1\u7684Agent\u52a0\u7801\uff0c\u90a3\u4e48\u8fd9\u7bc7\u6765\u81ea\u4e00\u7ebf\u5b9e\u6218\u7684\u7ecf\u9a8c\u603b\u7ed3\uff0c\u5efa\u8bae\u5148\u53cd\u590d\u9605\u8bfb\u80cc\u8bf5\uff5e<\/p>\n<p>\u4e8b\u60c5\u662f\u9171\u7d2b\u7684\u2014\u2014<\/p>\n<p>\u81ea\u4eceGLM-5\u53d1\u5e03\u4ee5\u6765\uff0c\u667a\u8c31\u901a\u8fc7\u89c2\u5bdf\u7528\u6237\u7684\u5927\u89c4\u6a21Coding Agent\u63a8\u7406\u8fc7\u7a0b\uff0c\u53d1\u73b0\u4e86\u4e09\u7c7b\u5f02\u5e38\u73b0\u8c61\uff1a<\/p>\n<p>\u8fd9\u5f15\u8d77\u4e86\u56e2\u961f\u5de5\u7a0b\u5e08\u7684\u8b66\u89c9\uff0c\u4e8e\u662f\u8bf4\u5e72\u5c31\u5e72\uff0c\u5148\u662f\u901a\u8fc7\u672c\u5730\u56de\u653e\u7528\u6237\u53cd\u9988\uff0c\u91cd\u590d\u8fd0\u884c\u76f8\u540c\u8bf7\u6c42\u6570\u767e\u6b21\uff0c\u7ed3\u679c\u59cb\u7ec8\u65e0\u6cd5\u89e6\u53d1\u5f02\u5e38\u3002<\/p>\n<p>\u6362\u8a00\u4e4b\uff0c\u6a21\u578b\u672c\u8eab\u5e76\u975e\u6839\u672c\u539f\u56e0\u3002<\/p>\n<p>\u5728\u8fdb\u4e00\u6b65\u6a21\u62df\u5728\u7ebf\u73af\u5883\u540e\uff0c\u56e2\u961f\u5c1d\u8bd5\u8c03\u6574PD\u5206\u79bb\u6bd4\u4f8b\u5e76\u6301\u7eed\u63d0\u9ad8\u7cfb\u7edf\u8d1f\u8f7d\uff0c\u5f02\u5e38\u73b0\u8c61\u7ec8\u4e8e\u5f97\u4ee5\u590d\u73b0\uff0c\u5728\u6bcf10000\u4e2a\u8bf7\u6c42\u4e2d\u5927\u7ea6\u80fd\u590d\u73b0\u51fa3-5\u4e2a\u5f02\u5e38\u8f93\u51fa\u3002<\/p>\n<p>\u8fd9\u8bf4\u660e\uff0c\u5f02\u5e38\u73b0\u8c61\u5f88\u6709\u53ef\u80fd\u51fa\u81ea\u9ad8\u8d1f\u8f7d\u4e0b\u7684\u63a8\u7406\u72b6\u6001\u7ba1\u7406\uff0c\u6307\u5411\u5e95\u5c42\u63a8\u7406\u94fe\u8def\u3002<\/p>\n<p>\u4f46\u540c\u65f6\u4e5f\u5f15\u51fa\u4e86\u53e6\u4e00\u4e2a\u95ee\u9898\uff0c\u7ebf\u4e0b\u7684\u590d\u73b0\u7387\u4ecd\u4f4e\u4e8e\u7528\u6237\u7ebf\u4e0a\u53cd\u9988\u7684\u9891\u7387\uff0c\u8fd9\u5c31\u610f\u5473\u7740\u73b0\u6709\u7684\u68c0\u6d4b\u65b9\u6cd5\u5b58\u5728\u9057\u6f0f\u6216\u89e6\u53d1\u6761\u4ef6\u5c1a\u672a\u5b8c\u5168\u8986\u76d6\u3002<\/p>\n<p>\u4e8e\u662f\u667a\u8c31\u56e2\u961f\u7ee7\u7eed\u5bf9\u5f02\u5e38\u8f93\u51fa\u7684\u68c0\u6d4b\u65b9\u6cd5\u8fdb\u884c\u4f18\u5316\u3002\u4ed6\u4eec\u53d1\u73b0\u6295\u673a\u91c7\u6837\uff08Speculative Decoding\uff09\u6307\u6807\u53ef\u4f5c\u4e3a\u5f02\u5e38\u68c0\u6d4b\u7684\u91cd\u8981\u53c2\u8003\u3002<\/p>\n<p>\u6295\u673a\u91c7\u6837\u539f\u672c\u7528\u4e8e\u63d0\u5347\u6a21\u578b\u63a8\u7406\u6027\u80fd\uff0c\u5b83\u5148\u7531\u5c0f\u6a21\u578b\u751f\u6210\u8349\u7a3f\uff08draft tokens\uff09\uff0c\u518d\u7531\u5927\u6a21\u578b\u9a8c\u8bc1\u662f\u5426\u63a5\u53d7\u8fd9\u4e9btoken\uff0c\u6700\u7ec8\u80fd\u591f\u5728\u4e0d\u6539\u53d8\u8f93\u51fa\u5206\u5e03\u7684\u60c5\u51b5\u4e0b\u63d0\u5347decode\u6548\u7387\u3002<\/p>\n<p>\u800c\u5728GLM-5\u7684\u4e09\u7c7b\u5f02\u5e38\u4e2d\uff0c\u4e71\u7801\u548c\u751f\u50fb\u5b57\u7684spec_accept_length\u975e\u5e38\u4f4e\uff0c\u4e5f\u5c31\u662f\u8bf4\u76ee\u6807\u6a21\u578b\u7684KV\u7f13\u5b58\u72b6\u6001\u4e0e\u8349\u7a3f\u6a21\u578b\u4e4b\u95f4\u5b58\u5728\u660e\u663e\u4e0d\u5339\u914d\u3002<\/p>\n<p>\u590d\u8bfb\u5219\u62e5\u6709\u8fc7\u9ad8\u7684spec_accept_length\uff0c\u8868\u660e\u635f\u574f\u7684KV\u7f13\u5b58\u53ef\u80fd\u5bfc\u81f4\u6ce8\u610f\u529b\u6a21\u5f0f\u9000\u5316\uff0c\u5c06\u751f\u6210\u8fc7\u7a0b\u63a8\u5411\u9ad8\u7f6e\u4fe1\u5ea6\u7684\u91cd\u590d\u5faa\u73af\u3002<\/p>\n<p>\u57fa\u4e8e\u4ee5\u4e0a\u89c2\u5bdf\uff0c\u667a\u8c31\u603b\u7ed3\u51fa\u4e86\u4e00\u5957\u5728\u7ebf\u5f02\u5e38\u76d1\u63a7\u7b56\u7565\uff1a<\/p>\n<p>\u5f53spec_accept_length\u6301\u7eed\u4f4e\u4e8e1.4\u4e14\u751f\u6210\u957f\u5ea6\u8d85\u8fc7128 token\uff0c\u6216\u8005spec_accept_rate\u8d85\u8fc70.96\uff0c\u7cfb\u7edf\u5c31\u4f1a\u4e3b\u52a8\u4e2d\u6b62\u5f53\u524d\u751f\u6210\uff0c\u5e76\u5c06\u8bf7\u6c42\u91cd\u65b0\u4ea4\u56de\u7ed9\u8d1f\u8f7d\u5747\u8861\u5668\u3002<\/p>\n<p>\u7d27\u63a5\u7740\uff0c\u667a\u8c31\u5f00\u59cb\u8fdb\u4e00\u6b65\u89e3\u6790\u5f02\u5e38\u539f\u56e0\uff1a<\/p>\n<p>PD\u5206\u79bb\u67b6\u6784\u4e0b\u7684KV Cache\u7ade\u6001<\/p>\n<p>\u56e2\u961f\u901a\u8fc7\u5206\u6790\u8bf7\u6c42\u751f\u547d\u5468\u671f\u548c\u63a8\u7406\u5f15\u64ce\u4e2d\u7684PD\u5206\u79bb\u6267\u884c\u65f6\u5e8f\uff0c\u5c06\u95ee\u9898\u5f52\u56e0\u4e8e\u8bf7\u6c42\u751f\u547d\u5468\u671f\u4e0eKV Cache\u56de\u6536\u4e0e\u590d\u7528\u65f6\u5e8f\u4e4b\u95f4\u7684\u4e0d\u4e00\u81f4\uff0c\u4ece\u800c\u5f15\u53d1\u7684KV Cache\u590d\u7528\u51b2\u7a81\u3002<\/p>\n<p>\u4e3a\u4e86\u6d88\u9664\u8fd9\u7c7b\u7ade\u6001\u60c5\u51b5\uff0c\u7814\u7a76\u4eba\u5458\u5728\u63a8\u7406\u5f15\u64ce\u4e2d\u5f15\u5165\u4e86\u66f4\u4e3a\u4e25\u683c\u7684\u65f6\u5e8f\u7ea6\u675f\uff0c\u4f1a\u5728\u8bf7\u6c42\u7ec8\u6b62\u548cKV Cache\u5199\u5165\u5b8c\u6210\u4e4b\u95f4\u5efa\u7acb\u663e\u5f0f\u540c\u6b65\u3002<\/p>\n<p>\u5177\u4f53\u6765\u8bf4\uff0c\u5728\u53d1\u51fa\u4e2d\u6b62\u6307\u4ee4\u540e\uff0c\u89e3\u7801\u9636\u6bb5\u4f1a\u5411\u9884\u586b\u5145\u9636\u6bb5\u53d1\u9001\u901a\u77e5\u3002\u9884\u586b\u5145\u9636\u6bb5\u53ea\u6709\u5728\u6ee1\u8db3\u4ee5\u4e0b\u4efb\u4e00\u6761\u4ef6\u65f6\u624d\u4f1a\u8fd4\u56de\u5b89\u5168\u56de\u6536\u4fe1\u53f7\uff1a\u672a\u542f\u52a8\u4efb\u4f55RDMA\u5199\u5165\uff0c\u6216\u6240\u6709\u5148\u524d\u53d1\u51fa\u7684\u5199\u5165\u64cd\u4f5c\u5df2\u5b8c\u5168\u5b8c\u6210\u3002\u800c\u89e3\u7801\u9636\u6bb5\u53ea\u6709\u5728\u6536\u5230\u6b64\u786e\u8ba4\u540e\u624d\u4f1a\u56de\u6536\u5e76\u91cd\u7528\u76f8\u5e94\u7684 KV Cache\u69fd\u4f4d\u3002<\/p>\n<p>\u8be5\u673a\u5236\u5c06\u786e\u4fddKV Cache\u5199\u5165\u4e0d\u4f1a\u8de8\u8d8a\u5185\u5b58\u590d\u7528\u8fb9\u754c\uff0c\u4ece\u800c\u907f\u514d\u8de8\u8bf7\u6c42\u7684KV Cache\u635f\u574f\u3002<\/p>\n<p>\u6700\u7ec8\u4fee\u590d\u8be5bug\u540e\uff0c\u5f02\u5e38\u8f93\u51fa\u7684\u53d1\u751f\u7387\u4ece\u7ea6\u4e07\u5206\u4e4b\u5341\u51e0\u4e0b\u964d\u81f3\u4e07\u5206\u4e4b\u4e09\u4ee5\u4e0b\u3002<\/p>\n<p>HiCache\u52a0\u8f7d\u65f6\u5e8f\u7f3a\u5931<\/p>\n<p>\u6b64\u5916\uff0c\u5f53KV Cache\u6362\u5165\u4e0e\u8ba1\u7b97\u91cd\u53e0\u65f6\uff0c\u5f53\u524d\u5b9e\u73b0\u672a\u80fd\u4fdd\u8bc1\u6570\u636e\u5728\u4f7f\u7528\u524d\u5df2\u5b8c\u6210\u52a0\u8f7d\uff0c\u5bfc\u81f4\u53ef\u80fd\u51fa\u73b0\u672a\u5c31\u7eeaKV Cache\u88ab\u8bbf\u95ee\u7684\u60c5\u51b5\u3002<\/p>\n<p>\u4e3a\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u56e2\u961f\u91cd\u6784\u4e86HiCache\u8bfb\u53d6\u6d41\u7a0b\uff0c\u540c\u65f6\u5f15\u5165\u6570\u636e\u52a0\u8f7d\u4e0e\u8ba1\u7b97\u4e4b\u95f4\u7684\u663e\u5f0f\u540c\u6b65\u7ea6\u675f\u3002<\/p>\n<p>\u5728\u542f\u52a8Indexer\u7b97\u5b50\u4e4b\u524d\uff0c\u5148\u63d2\u5165\u4e00\u4e2aLoad Stream\u540c\u6b65\u70b9\uff0c\u786e\u4fdd\u76f8\u5e94\u7ea7\u522b\u7684Indexer\u7f13\u5b58\u5df2\u5b8c\u5168\u52a0\u8f7d\u3002Forward Stream\u53ea\u6709\u5728\u6570\u636e\u51c6\u5907\u5c31\u7eea\u540e\u624d\u4f1a\u8fdb\u884c\u8ba1\u7b97\uff0c\u4ece\u800c\u6d88\u9664\u4e86read-before-ready\u7684\u95ee\u9898\u3002<\/p>\n<p>\u5e94\u7528\u6b64\u4fee\u590d\u540e\uff0c\u5728\u76f8\u540c\u7684\u5de5\u4f5c\u8d1f\u8f7d\u6761\u4ef6\u4e0b\uff0c\u7531\u6267\u884c\u65f6\u5e8f\u4e0d\u4e00\u81f4\u5f15\u8d77\u7684\u5f02\u5e38\u88ab\u6d88\u9664\uff0c\u7cfb\u7edf\u7ec8\u4e8e\u5f97\u4ee5\u7a33\u5b9a\u3002<\/p>\n<p>\u4e8b\u5b9e\u4e0a\uff0c\u8fd9\u4e24\u79cdBug\u90fd\u6307\u5411\u4e86\u540c\u4e00\u4e2a\u5e38\u89c1\u7684\u7cfb\u7edf\u74f6\u9888\uff1a<\/p>\n<p>\u5728\u957f\u4e0a\u4e0b\u6587\u7684Coding Agent Serving\u4efb\u52a1\u4e2d\uff0cPrefill\u9636\u6bb5\u5df2\u7ecf\u6210\u4e3a\u5f71\u54cd\u7cfb\u7edf\u6027\u80fd\u7684\u4e3b\u8981\u56e0\u7d20\u3002<\/p>\n<p>\u4e8e\u662f\u4e3a\u4e86\u7f13\u89e3Prefill\u9636\u6bb5\u5728\u9ad8\u5e76\u53d1\u4e0b\u7684\u5185\u5b58\u548c\u5e26\u5bbd\u538b\u529b\uff0c\u56e2\u961f\u53e6\u5916\u8bbe\u8ba1\u4e86KV Cache\u5206\u5c42\u5b58\u50a8\u65b9\u6848\u2014\u2014LayerSplit\u3002<\/p>\n<p>\u5728\u8be5\u65b9\u6848\u4e2d\uff0c\u6bcf\u4e2aGPU\u53ea\u5b58\u50a8\u90e8\u5206\u5c42\u7684KV Cache\uff0c\u663e\u8457\u964d\u4f4e\u4e86\u6bcf\u4e2aGPU\u7684\u5185\u5b58\u5360\u7528\u3002\u7136\u540e\u5728\u6267\u884cAttention\u8ba1\u7b97\u524d\uff0c\u5c06\u5bf9\u5e94\u5c42\u7684KV Cache\u5e7f\u64ad\u7ed9\u5176\u4ed6\u76f8\u5173rank\u3002<\/p>\n<p>\u4e3a\u4e86\u964d\u4f4e\u901a\u4fe1\u5f00\u9500\uff0c\u8fd8\u8fdb\u4e00\u6b65\u8bbe\u8ba1\u6709KV Cache\u5e7f\u64ad\u4e0eindexer\u8ba1\u7b97\u7684\u91cd\u53e0\u673a\u5236\uff0c\u5c06\u901a\u4fe1\u5ef6\u8fdf\u9690\u85cf\u5728\u8ba1\u7b97\u8fc7\u7a0b\u4e2d\u3002\u8fd9\u6837\u552f\u4e00\u7684\u989d\u5916\u901a\u4fe1\u5f00\u9500\u5c31\u6765\u81eaIndexer Cache\u7684\u5e7f\u64ad\uff0c\u5176\u5927\u5c0f\u4ec5\u4e3aKV Cache\u7684\u516b\u5206\u4e4b\u4e00\uff0c\u6574\u4f53\u901a\u4fe1\u6210\u672c\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8ba1\u3002<\/p>\n<p>\u56e2\u961f\u5c06LayerSplit\u548cGLM-5.1\u7ed3\u5408\u53d1\u73b0\uff0c\u5728Cache\u547d\u4e2d\u7387\u8fbe\u523090%\u3001\u8bf7\u6c42\u957f\u5ea6\u572840k\u5230120k\u533a\u95f4\u5185\u65f6\uff0c\u7cfb\u7edf\u541e\u5410\u91cf\u63d0\u9ad8\u4e8610%\u5230132%\uff0c\u4e14\u968f\u7740\u4e0a\u4e0b\u6587\u957f\u5ea6\u7684\u589e\u52a0\uff0c\u6536\u76ca\u4e5f\u968f\u4e4b\u589e\u957f\u3002<\/p>\n<p>\u603b\u4f53\u800c\u8a00\uff0c\u8be5\u4f18\u5316\u663e\u8457\u63d0\u5347\u4e86\u7cfb\u7edf\u5728Coding Agent\u573a\u666f\u4e0b\u7684\u5904\u7406\u80fd\u529b\u3002<\/p>\n<p>\u540c\u65f6\u667a\u8c31\u4e5f\u8ba4\u4e3a\uff0c\u5f53\u667a\u80fd\u771f\u6b63\u8fdb\u5165\u9ad8\u5e76\u53d1\u3001\u957f\u4e0a\u4e0b\u6587\u7684Coding Agent\u573a\u666f\u540e\uff0c\u7ef4\u62a4\u63a8\u7406\u57fa\u7840\u8bbe\u65bd\u7684\u8f93\u51fa\u8d28\u91cf\u53d8\u5f97\u81f3\u5173\u91cd\u8981\u3002\u672a\u6765\u5927\u89c4\u6a21AI\u9700\u8981\u7684\u4e0d\u4ec5\u662fScaling Law\u63a8\u52a8\u7684\u80fd\u529b\u589e\u957f\uff0c\u8fd8\u5fc5\u987b\u6709\u7b49\u91cf\u7ea7\u7684\u7cfb\u7edf\u5de5\u7a0b\u652f\u6491\u3002<\/p>\n<p>\u53c2\u8003\u94fe\u63a5\uff1a[1]https:\/\/z.ai\/blog\/scaling-pain[2]https:\/\/www.zhipuai.cn\/zh\/research\/159<\/p>\n<blockquote>\n<p><strong>\u7248\u6743\u58f0\u660e\uff1a<\/strong><\/p>\n<p>\u672c\u535a\u5ba2\u90e8\u5206\u5185\u5bb9\u4e3a\u8f6c\u8f7d\u6587\u7ae0\uff0c\u65e8\u5728\u5206\u4eab\u6709\u4ef7\u503c\u7684\u4fe1\u606f\uff0c\u7248\u6743\u5f52\u539f\u4f5c\u8005\u6240\u6709\u3002<\/p>\n<p>\u8f6c\u8f7d\u4ec5\u4e3a\u4e2a\u4eba\u5b66\u4e60\u4e0e\u4ea4\u6d41\u76ee\u7684\uff0c\u4e0d\u5bf9\u6587\u7ae0\u89c2\u70b9\u8d1f\u8d23\uff0c\u4ea6\u4e0d\u7528\u4e8e\u4efb\u4f55\u5546\u4e1a\u7528\u9014\u3002<\/p>\n<p>\u5982\u6d89\u53ca\u7248\u6743\u95ee\u9898\uff0c\u8bf7\u8054\u7cfb\u672c\u4eba\u5220\u9664\u3002<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"<p>\u9e6d\u7fbd \u53d1\u81ea \u51f9\u975e\u5bfa<\/p>\n<p>Scaling\u5373\u6b63\u4e49\uff1f\u667a\u8c31\u6320\u4e86\u6320\u5934\u2014\u2014<\/p>\n<p>\u5f88\u75db\u82e6\uff0c\u800c\u4e14\u538b\u529b\u5c71\u5927\u2026\u2026<\/p>\n<p>\u667a\u8c31\u6700\u65b0\u53d1\u5e03\u7684\u4e00\u7bc7\u6280\u672f\u535a\u5ba2\uff0c\u753b\u98ce\u7a0d\u5fae\u6709\u70b9\u4e0d\u4e00\u6837\uff1a<\/p>\n<p>\u6ca1\u6709\u8fc7\u53bb\u7684\u786c\u6838\u6280\u672f\u8f93\u51fa\uff0c\u53cd\u800c\u5927\u5012\u82e6\u6c34\u4eceGLM-5\u4ee5\u6765\u7684\u5404\u79cd\u82b1\u5f0f\u8e29\u5751\uff0c\u5b98\u65b9\u79f0\u4e4b\u4e3a\u300cScaling Pain\u300d\u3002<\/p>\n<p>\u6211\u4eec\u7684\u63a8\u7406\u57fa\u7840\u8bbe\u65bd\u6b63\u627f\u53d7\u7740\u524d\u6240\u672a\u6709\u7684\u538b\u529b\uff0c\u6bcf\u5929\u90fd\u8981\u670d\u52a1\u6570\u4ebf\u6b21Coding Agent\u8c03\u7528\u3002<\/p>\n<p>\u8fc7\u53bb\u51e0\u5468\uff0c\u4e00\u4e9b\u7528\u6237\u5728\u4f7f\u7528GLM-5\u7cfb\u5217\u6a21\u578b\u6267\u884c\u590d\u6742<\/p>\n","protected":false},"author":1,"featured_media":4545,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[450,23],"class_list":["post-4549","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-xwzx","tag-glm-5","tag-23"],"_links":{"self":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/4549","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/comments?post=4549"}],"version-history":[{"count":0,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/posts\/4549\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media\/4545"}],"wp:attachment":[{"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/media?parent=4549"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/categories?post=4549"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/1ow.top\/index.php\/wp-json\/wp\/v2\/tags?post=4549"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}