{"id":5094,"date":"2026-01-07T15:48:19","date_gmt":"2026-01-07T07:48:19","guid":{"rendered":"https:\/\/www.haruhi.fans\/?p=5094"},"modified":"2026-01-07T16:46:56","modified_gmt":"2026-01-07T08:46:56","slug":"%e6%95%b0%e6%8d%ae%e9%9b%86%e7%ad%9b%e9%80%89","status":"publish","type":"post","link":"https:\/\/www.haruhi.fans\/?p=5094","title":{"rendered":"\u6570\u636e\u96c6\u7b5b\u9009"},"content":{"rendered":"\n<div data-wp-interactive=\"core\/file\" class=\"wp-block-file\"><object data-wp-bind--hidden=\"!state.hasPdfPreview\" hidden class=\"wp-block-file__embed\" data=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/\u7ec4\u4f1a-1.7.pdf\" type=\"application\/pdf\" style=\"width:100%;height:600px\" aria-label=\"\u5d4c\u5165 \u7ec4\u4f1a-1.7\"><\/object><a id=\"wp-block-file--media-09db6653-9931-4877-989b-84d8d8f1bdc1\" href=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/\u7ec4\u4f1a-1.7.pdf\">\u7ec4\u4f1a-1.7<\/a><a href=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/\u7ec4\u4f1a-1.7.pdf\" class=\"wp-block-file__button wp-element-button\" download aria-describedby=\"wp-block-file--media-09db6653-9931-4877-989b-84d8d8f1bdc1\">\u4e0b\u8f7d<\/a><\/div>\n\n\n\n<h2 class=\"wp-block-heading\">tinyBenchmarks: evaluating LLMs with fewer examples<\/h2>\n\n\n\n<p>\u7eaf\u6570\u5b66\u65b9\u6cd5\uff0c\u805a\u7c7b<\/p>\n\n\n\n<p>\u4e09\u6b65\u8d70\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>IRT \u53c2\u6570\u5316<\/strong>\uff1a\u9996\u5148\u5229\u7528\u5386\u53f2\u6570\u636e\uff0c\u901a\u8fc7 IRT \u6a21\u578b\u5b66\u4e60\u5168\u91cf\u6570\u636e\u96c6\u4e2d\u6bcf\u4e2a\u95ee\u9898\u7684\u6f5c\u5728\u53c2\u6570\uff08\u4e3b\u8981\u662f<strong>\u533a\u5206\u5ea6<\/strong> $\\alpha$ \u548c <strong>\u96be\u5ea6<\/strong> $\\beta$\uff09\u3002<\/li>\n\n\n\n<li><strong>\u805a\u7c7b\u62bd\u6837<\/strong>\uff1a\u5c06\u8fd9\u4e9b\u53c2\u6570\u4f5c\u4e3a\u6bcf\u4e2a\u95ee\u9898\u7684\u7279\u5f81\u5411\u91cf\u8fdb\u884c\u805a\u7c7b\uff08Clustering\uff09\u3002<\/li>\n\n\n\n<li><strong>\u9009\u51fa\u4ee3\u8868<\/strong>\uff1a\u8fd9\u6837\u9009\u51fa\u7684 100 \u4e2a\u4f8b\u5b50\u80fd\u6700\u5927\u9650\u5ea6\u5730\u8986\u76d6\u4e0d\u540c\u96be\u5ea6\u548c\u533a\u5206\u5ea6\u7684\u9898\u76ee\uff0c\u907f\u514d\u4e86\u968f\u673a\u62bd\u6837\u7684\u504f\u5dee\u3002<\/li>\n<\/ul>\n\n\n\n<p>IRT\u5982\u4f55\u4f30\u7b97\u5462\uff1f<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u90e8\u5206 A\uff1a\u89c2\u6d4b\u5206\uff08\u4f4e\u504f\u5dee\uff0c\u9ad8\u65b9\u5dee\uff09<\/strong> \u76f4\u63a5\u8ba1\u7b97\u6a21\u578b\u5728\u90a3 100 \u4e2a\u201c\u951a\u70b9\u201d\u95ee\u9898\u4e0a\u7684\u52a0\u6743\u5e73\u5747\u5206\u3002\u8fd9\u53cd\u6620\u4e86\u5f53\u4e0b\u7684\u771f\u5b9e\u8868\u73b0\uff0c\u4f46\u6837\u672c\u5c11\u5bfc\u81f4\u6ce2\u52a8\u5927\u3002<\/li>\n\n\n\n<li><strong>\u90e8\u5206 B\uff1a\u9884\u6d4b\u5206\uff08\u4f4e\u65b9\u5dee\uff0c\u6f5c\u5728\u504f\u5dee\uff09<\/strong> \u5229\u7528 IRT \u6a21\u578b\u3002\u6839\u636e\u6a21\u578b\u5bf9\u8fd9 100 \u9898\u7684\u56de\u7b54\uff0c\u53cd\u5411\u63a8\u65ad\u51fa\u8be5\u6a21\u578b\u7684**\u201c\u6f5c\u5728\u80fd\u529b\u503c\u201d ($\\theta$)<strong>\u3002\u7136\u540e\u628a\u8fd9\u4e2a\u80fd\u529b\u503c\u4ee3\u5165 IRT \u516c\u5f0f\uff0c\u53bb<\/strong>\u9884\u6d4b**\u8be5\u6a21\u578b\u5728\u5269\u4f59\u6240\u6709\u672a\u6d4b\u8bd5\u9898\u76ee\u4e0a\u7684\u5f97\u5206\u6982\u7387\u3002<\/li>\n\n\n\n<li><strong>\u6df7\u5408 (IRT++)<\/strong> \u6700\u7ec8\u5f97\u5206 = $\\lambda \\times$ <strong>\u89c2\u6d4b\u5206<\/strong> + $(1-\\lambda) \\times$ <strong>\u9884\u6d4b\u5206<\/strong>\u3002\n<ul class=\"wp-block-list\">\n<li>$\\lambda$ \u662f\u4e00\u4e2a\u901a\u8fc7\u7edf\u8ba1\u5b66\u65b9\u6cd5\u8ba1\u7b97\u51fa\u7684\u6743\u91cd\uff0c\u7528\u4e8e\u5728\u65b9\u5dee\u548c\u504f\u5dee\u4e4b\u95f4\u53d6\u5f97\u6700\u4f73\u5e73\u8861\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>Open LLM Leaderboard (\u5927\u89c4\u6a21\u6570\u636e)<\/strong>\n<ul class=\"wp-block-list\">\n<li><strong>\u603b\u6a21\u578b\u6570<\/strong>\uff1a\u7ea6\u00a0<strong>400 \u4e2a<\/strong>\u00a0(\u5177\u4f53\u4e3a 395 \u4e2a + 40 \u4e2a\u7279\u5b9a\u5fae\u8c03\u6a21\u578b)\u3002<\/li>\n\n\n\n<li><strong>\u8bad\u7ec3\u96c6 (Train)<\/strong>\uff1a\u7ea6\u00a0<strong>300+ \u4e2a<\/strong>\uff08\u8f83\u65e7\u7684\u6a21\u578b\uff09\u3002<\/li>\n\n\n\n<li><strong>\u6d4b\u8bd5\u96c6 (Test)<\/strong>\uff1a\u7ea6\u00a0<strong>40-80 \u4e2a<\/strong>\uff08\u6700\u65b0\u53d1\u5e03\u7684\u6a21\u578b\uff09\u3002<\/li>\n\n\n\n<li><em>\u6ce8\uff1a\u8fd9\u662f\u6570\u636e\u6700\u4e30\u5bcc\u7684\u4e00\u7ec4\uff0c\u9a8c\u8bc1\u6548\u679c\u4e5f\u6700\u5f3a\u3002<\/em><\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>HELM (\u5c0f\u89c4\u6a21\u6570\u636e)<\/strong>\n<ul class=\"wp-block-list\">\n<li><strong>\u603b\u6a21\u578b\u6570<\/strong>\uff1a\u4ec5\u00a0<strong>37 \u4e2a<\/strong>\u3002<\/li>\n\n\n\n<li><strong>\u8bad\u7ec3\u96c6<\/strong>\uff1a\u7ea6\u00a0<strong>19-28 \u4e2a<\/strong>\uff0850% &#8211; 75%\uff09\u3002<\/li>\n\n\n\n<li><strong>\u6d4b\u8bd5\u96c6<\/strong>\uff1a\u7ea6\u00a0<strong>9-18 \u4e2a<\/strong>\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>MMLU \/ AlpacaEval<\/strong>\n<ul class=\"wp-block-list\">\n<li>\u5927\u81f4\u90fd\u5728\u00a0<strong>100-300 \u4e2a<\/strong>\u6a21\u578b\u8fd9\u4e2a\u91cf\u7ea7\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<p>\u6570\u636e\u89c4\u6a21<\/p>\n\n\n\n<p>\u4f5c\u8005\u8ba4\u4e3a100\u5de6\u53f3\u5c31\u6bd4\u8f83\u597d\u4e86\uff0c\u53ef\u4ee5\u8fbe\u52302%\u4ee5\u5185\u7684\u8bef\u5dee\u63a7\u5236<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th><strong>\u57fa\u51c6\u6d4b\u8bd5 (Benchmark)<\/strong><\/th><th><strong>\u539f\u59cb\u9898\u91cf (Original)<\/strong><\/th><th><strong>\u7f29\u51cf\u540e\u9898\u91cf (Tiny)<\/strong><\/th><th><strong>\u538b\u7f29\u6bd4\u4f8b<\/strong><\/th><th><strong>\u5907\u6ce8<\/strong><\/th><\/tr><\/thead><tbody><tr><td><strong>MMLU<\/strong><\/td><td><strong>14,042<\/strong><\/td><td><strong>100<\/strong><\/td><td><strong>0.7%<\/strong><\/td><td>\u538b\u7f29\u4e86140\u500d<\/td><\/tr><tr><td><strong>Open LLM LB<\/strong><\/td><td><strong>~29,000<\/strong><\/td><td><strong>600<\/strong><\/td><td><strong>2%<\/strong><\/td><td>\u542b6\u4e2a\u4efb\u52a1\uff0c\u6bcf\u4e2a\u4efb\u52a1\u7559100\u9898<\/td><\/tr><tr><td><strong>HELM<\/strong><\/td><td><strong>~5,000<\/strong><\/td><td><strong>100<\/strong><\/td><td><strong>2%<\/strong><\/td><td>\u6bcf\u4e2a\u573a\u666f(Scenario)\u7559100\u9898<\/td><\/tr><tr><td><strong>AlpacaEval 2.0<\/strong><\/td><td><strong>805<\/strong><\/td><td><strong>100<\/strong><\/td><td><strong>12%<\/strong><\/td><td>\u8fd9\u662f\u4e00\u4e2a\u6bd4\u8f83\u5c0f\u7684\u57fa\u51c6<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">metabench &#8212; A Sparse Benchmark of Reasoning and Knowledge in Large Language Models<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"790\" height=\"548\" data-original=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-5105\"  sizes=\"auto, (max-width: 790px) 100vw, 790px\" \/><\/div><\/figure>\n\n\n\n<p>1.<strong>\u6570\u636e\u9884\u5904\u7406<\/strong><\/p>\n\n\n\n<p>\u53bb\u6389\u592a\u7b80\u5355\u592a\u96be\u7684\uff0c\u65b9\u5dee\u8fc7\u9ad8\u7684\u9898\u76ee\u3002 \u8ba1\u7b97<strong>\u90e8\u5206-\u6574\u4f53\u76f8\u5173\u6027 (Point-biserial correlation)<\/strong>\uff1a\u5254\u9664\u90a3\u4e9b\u4e0e\u603b\u5206\u76f8\u5173\u6027\u6781\u4f4e\u751a\u81f3\u8d1f\u76f8\u5173\u7684\u9898\u76ee\uff08\u5373\u7b54\u5bf9\u8be5\u9898\u7684\u6a21\u578b\u53cd\u800c\u603b\u5206\u8f83\u4f4e\u7684\u5f02\u5e38\u9898\uff09\u3002<\/p>\n\n\n\n<p>2.<strong>\u5fc3\u7406\u6d4b\u91cf\u5b66\u5efa\u6a21\uff1a\u9879\u76ee\u53cd\u5e94\u7406\u8bba (Item Response Theory, IRT)<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5047\u8bbe\uff1a<\/strong> \u6bcf\u4e2a\u6a21\u578b\u6709\u4e00\u4e2a\u6f5c\u5728\u80fd\u529b\u503c $\uff08\\theta\uff0cscalar ability\uff09$\uff0c\u6bcf\u4e2a\u9898\u76ee\u6709\u5176\u7279\u6027\u53c2\u6570\uff08\u5982\u96be\u5ea6 $\\delta$ \u548c\u533a\u5206\u5ea6 $a$\uff09\u3002<\/li>\n\n\n\n<li><strong>\u6a21\u578b\uff1a<\/strong> \u4f5c\u8005\u4f7f\u7528\u4e86 2PL\u30013PL \u548c 4PL\uff08\u53c2\u6570\u903b\u8f91\u65af\u8482\uff09\u6a21\u578b\u6765\u62df\u5408\u6570\u636e\u3002\n<ul class=\"wp-block-list\">\n<li>\u4f8b\u5982 2PL \u6a21\u578b\u516c\u5f0f\uff1a$P(\\text{correct}) = \\sigma(a_i \\theta_j &#8211; \\delta_i)$\u3002\u8fd9\u610f\u5473\u7740\u6a21\u578b\u7b54\u5bf9\u9898\u76ee\u7684\u6982\u7387\u53d6\u51b3\u4e8e\u5176\u80fd\u529b\u503c<strong>\u4e0e\u9898\u76ee\u96be\u5ea6\u7684\u5173\u7cfb\uff0c\u4ee5\u53ca\u9898\u76ee\u7684\u533a\u5206\u5ea6<\/strong>\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<p>3.<strong>\u4fe1\u606f\u8fc7\u6ee4 (Information Filtering)<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8d39\u96ea\u4fe1\u606f\u91cf (Fisher Information)\uff1a<\/strong>\u00a0\u4f5c\u8005\u8ba1\u7b97\u4e86\u6bcf\u4e2a\u9898\u76ee\u5728\u4e0d\u540c\u80fd\u529b\u6c34\u5e73\u4e0b\u7684\u4fe1\u606f\u91cf\u3002<\/li>\n\n\n\n<li><strong>\u7b5b\u9009\u7b56\u7565\uff1a<\/strong>\u00a0\u5e76\u6ca1\u6709\u968f\u673a\u62bd\u9898\uff0c\u800c\u662f\u9009\u62e9\u90a3\u4e9b\u80fd\u63d0\u4f9b\u6700\u5927<strong>\u8d39\u96ea\u4fe1\u606f\u91cf<\/strong>\u7684\u9898\u76ee\u3002\u8fd9\u610f\u5473\u7740\u9009\u51fa\u7684\u9898\u76ee\u6700\u80fd\u533a\u5206\u6a21\u578b\u4e4b\u95f4\u7684\u7ec6\u5fae\u80fd\u529b\u5dee\u5f02\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\"><a href=\"https:\/\/aclanthology.org\/2025.acl-long.1477.pdf\">SubLIME: Subset Selection via Rank Correlation Prediction for Data-Efficient\u00a0LLM\u00a0Evaluation<\/a><\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image-1-1024x763.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"763\" data-original=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image-1-1024x763.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-5106\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p><strong>\u7279\u5f81\u63d0\u53d6 -&gt; \u6a21\u578b\u70ed\u8eab -&gt; \u9884\u6d4b\u8bad\u7ec3 -&gt; \u9a8c\u8bc1\u5bf9\u6bd4<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u7279\u5f81\u63d0\u53d6<\/strong>\u7684\u65f6\u5019\uff0c\u501f\u52a9llm\u6a21\u578b\u5bf9\u6570\u636e\u96c6\u6bcf\u4e00\u9053\u9898\u8ba1\u7b97\u4e09\u4e2a\u6307\u6807\uff0c\u96be\u5ea6\uff0c\u8d28\u91cf\uff0c\u5206\u5e03\u7b49<\/li>\n\n\n\n<li><strong>\u6a21\u578b\u70ed\u8eab\u9636\u6bb5<\/strong>\u6765\u8bf4\uff0c\u6ca1\u6709\u8ba9\u5168\u90e8\u6a21\u578b\u8dd1\u5168\u91cf\u6d4b\u8bd5\uff0c\u800c\u662f\u6311\u9009\u4e86\u51e0\u4e2a\u951a\u70b9\u6a21\u578b\uff0c\u7528\u8fd9\u4e9b\u951a\u70b9\u6a21\u578b\u6765\u8dd1\u5b8c\u6574\u6570\u636e\u96c6\uff0c\u7136\u540e\u7528\u8fd9\u51e0\u4e2a\u6a21\u578b\u7684\u6392\u540d\u4f5c\u4e3a\u6807\u51c6\u7b54\u6848\u6765\u8bad\u7ec3<\/li>\n\n\n\n<li><strong>\u8bad\u7ec3\u6392\u540d\u9884\u6d4b\u6a21\u578b<\/strong>\uff0c\u8bad\u7ec3\u4e86\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\uff0c\u8f93\u5165\u5404\u4e2a\u5b50\u96c6\u7684\u7279\u5f81+\u951a\u70b9\u7684\u8868\u73b0\u3002\u9884\u6d4b\u8fd9\u4e9b\u6392\u540d\u548c\u5168\u91cf\u771f\u5b9e\u6392\u540d\u591a\u50cf<\/li>\n\n\n\n<li><strong>\u9a8c\u8bc1\u4e0e\u5bf9\u6bd4<\/strong> \u5bf9\u4e8e\u672a\u77e5\u7684\u6a21\u578b\u6d4b\u8bd5\u3002\u770b\u5b50\u96c6\u548c\u7528\u5168\u96c6\u6d4b\u51fa\u6765\u7684\u6392\u540d\u7684<strong>\u65af\u76ae\u5c14\u66fc\u7b49\u7ea7\u76f8\u5173\u7cfb\u6570<\/strong>\uff0c\u8d8a\u9ad8\u8d8a\u51c6\u3002<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">Rethinking LLM Evaluation: Can We Evaluate LLMs with 200x Less Data?<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image-2-1024x800.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"800\" data-original=\"https:\/\/www.haruhi.fans\/wp-content\/uploads\/2026\/01\/image-2-1024x800.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-5107\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p>1.\u6b65\u9aa41\uff0c\u521d\u7b5b\uff0c\u7b5b\u9009\u6389\u9009\u62e9\u9ad8\u5ea6\u91cd\u590d\u7684\u6837\u672c\u3002\u5229\u7528<strong>embadding\u76f8\u4f3c\u5ea6\uff08\u8bed\u4e49\uff09\u548c\u6392\u540d\u76f8\u5173\u6027\uff08\u6548\u679c\uff09<\/strong>\uff0c\u8d85\u8fc7\u4e00\u5b9a\u9608\u503c\u7684\u8bdd\u5254\u9664\uff0c\u53ea\u4fdd\u7559\u4e00\u4e2a<\/p>\n\n\n\n<p>2.\u6b65\u9aa42\uff0c\u4f7f\u7528\u9057\u4f20\u7b97\u6cd5\uff0c\u9009\u51fa\u7279\u5b9a\u7684\u81ea\u5df1\u9884\u6d4b\u5728\u6837\u672c\u4e0a\u7684\u5f97\u5206\uff0c\u8bef\u5dee\u8d8a\u5c0f\u8d8a\u597d \u2022 <strong>\u76ee\u7684\uff1a<\/strong>&nbsp;\u4ece\u8fc7\u6ee4\u540e\u7684\u6570\u636e\u4e2d\u641c\u7d22\u51fa\u6700\u4f73\u7ec4\u5408\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5c06\u201c\u6311\u9009\u5b50\u96c6\u201d\u5efa\u6a21\u4e3a\u4f18\u5316\u95ee\u9898\u3002\u4f7f\u7528<strong>\u9057\u4f20\u7b97\u6cd5 (Genetic Algorithm)<\/strong>\u00a0\u8fdb\u884c\u8fed\u4ee3\u641c\u7d22\uff0c\u76ee\u6807\u51fd\u6570\u662f\u6700\u5c0f\u5316\u201c\u5b50\u96c6\u5f97\u5206\u201d\u4e0e\u201c\u5168\u96c6\u5f97\u5206\u201d\u4e4b\u95f4\u7684<strong>\u9884\u6d4b\u8bef\u5dee (RMSE)<\/strong>\u3002\n<ol class=\"wp-block-list\">\n<li>\u968f\u673a\u751f\u6210\u591a\u4e2a\u5b50\u96c6\uff08\u79cd\u7fa4\uff09\u3002<\/li>\n\n\n\n<li>\u8bc4\u4f30\u6bcf\u4e2a\u5b50\u96c6\u7684\u9884\u6d4b\u80fd\u529b\u3002<\/li>\n\n\n\n<li>\u901a\u8fc7<strong>\u9526\u6807\u8d5b\u9009\u62e9\u3001\u4ea4\u53c9 (Crossover)\u3001\u53d8\u5f02 (Mutation)<\/strong>\u00a0\u751f\u6210\u65b0\u4e00\u4ee3\u5b50\u96c6\u3002<\/li>\n\n\n\n<li>\u4e0d\u65ad\u8fed\u4ee3\uff0c\u76f4\u5230\u627e\u5230\u8868\u73b0\u6700\u597d\u7684\u5b50\u96c6\u3002<\/li>\n<\/ol>\n<\/li>\n<\/ul>\n\n\n\n<p>3.\u6b65\u9aa43\uff1a\u57fa\u4e8e\u5f52\u56e0\u7684\u6837\u672c\u7cbe\u9009<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u76ee\u7684\uff1a<\/strong>\u00a0\u89e3\u51b3\u5355\u7eaf\u9057\u4f20\u7b97\u6cd5\u53ef\u80fd\u9677\u5165\u5c40\u90e8\u6700\u4f18\u7684\u95ee\u9898\uff0c\u63d0\u9ad8\u641c\u7d22\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u624b\u6bb5\uff1a<\/strong>\u00a0\u8ba1\u7b97\u6bcf\u4e2a\u6837\u672c\u5bf9\u9884\u6d4b\u603b\u5206\u7684<strong>\u8d21\u732e\u5ea6 (Attribution)<\/strong>\u3002\u5c06\u6837\u672c\u5206\u4e3a\u201c\u9ad8\u8d21\u732e\u201d\u3001\u201c\u4f4e\u8d21\u732e\u201d\u548c\u201c\u968f\u673a\u201d\u4e09\u7ec4\uff0c\u5728\u8fd9\u4e9b\u7ec4\u5185\u518d\u6b21\u8fd0\u884c\u9057\u4f20\u7b97\u6cd5\u8fdb\u884c\u7cbe\u7ec6\u5316\u641c\u7d22\uff0c\u6700\u7ec8\u9009\u51fa\u65e2\u5177\u4ee3\u8868\u6027\u53c8\u5177\u591a\u6837\u6027\u7684\u6d4b\u8bd5\u96c6\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>tinyBenchmarks: evaluating LLMs with fewer examples \u7eaf\u6570\u5b66 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[],"class_list":["post-5094","post","type-post","status-publish","format-standard","hentry","category-note"],"_links":{"self":[{"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/posts\/5094","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5094"}],"version-history":[{"count":4,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/posts\/5094\/revisions"}],"predecessor-version":[{"id":5108,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=\/wp\/v2\/posts\/5094\/revisions\/5108"}],"wp:attachment":[{"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5094"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5094"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.haruhi.fans\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5094"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}