{"id":98590,"date":"2026-09-01T07:00:23","date_gmt":"2026-08-31T23:00:23","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/98590.html"},"modified":"2026-09-01T07:00:23","modified_gmt":"2026-08-31T23:00:23","slug":"27%e5%b1%8a%e5%a4%a7%e6%a8%a1%e5%9e%8b%e9%9d%a2%e8%af%95%e5%87%86%e5%a4%87%ef%bc%88%e4%ba%94%e5%8d%81%e4%b8%83%ef%bc%89%ef%bc%9a%e5%a4%9a%e6%a8%a1%e6%80%81%e5%a4%a7%e6%a8%a1%e5%9e%8b%e8%af%84%e6%b5%8b","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/98590.html","title":{"rendered":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4"},"content":{"rendered":"<h2>27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907&#xff08;\u4e94\u5341\u4e03&#xff09;&#xff1a;\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4<\/h2>\n<h3>\u5f15\u8a00&#xff1a;\u4e0e\u4e0a\u4e00\u7bc7\u3001\u7cfb\u5217\u7684\u5173\u7cfb<\/h3>\n<p>\u4e0a\u4e00\u7bc7&#xff08;\u4e94\u5341\u516d&#xff09;\u6211\u4eec\u804a\u4e86\u300c\u5927\u6a21\u578b\u8bad\u63a8\u4e00\u4f53\u5316\u5de5\u7a0b\u300d&#xff0c;\u628a\u8bad\u7ec3\u4ea7\u7269\u7a33\u5b9a\u5730\u4ea4\u4ed8\u5230\u63a8\u7406\u4fa7&#xff1b;\u4f46\u8fd9\u53ea\u89e3\u51b3\u4e86\u4e00\u534a\u95ee\u9898\u2014\u2014\u4f60\u4ea4\u4ed8\u7684\u6a21\u578b\u5230\u5e95\u597d\u4e0d\u597d\u3001\u9000\u4e86\u591a\u5c11\u3001\u54ea\u7c7b\u6837\u672c\u5d29\u4e86&#xff0c;\u5fc5\u987b\u9760\u4e00\u5957\u8bc4\u6d4b\u5de5\u7a0b\u6765\u56de\u7b54\u3002\u5f80\u524d\u770b&#xff0c;&#xff08;\u4e94\u5341\u4e09&#xff09;\u8bb2\u4e86\u591a\u6a21\u6001\u63a8\u7406\u670d\u52a1\u5316\u3001&#xff08;\u4e94\u5341\u4e94&#xff09;\u8bb2\u4e86\u591a\u6a21\u6001 RAG \u5de5\u7a0b&#xff0c;\u672c\u7bc7\u628a\u5b83\u4eec\u6536\u53e3\u5230\u4e00\u4e2a\u5e38\u88ab\u5ffd\u89c6\u5374\u6709\u6781\u9ad8\u9762\u8bd5\u6743\u91cd\u7684\u73af\u8282&#xff1a;\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u3002<\/p>\n<p>\u4e3a\u4ec0\u4e48\u9762\u8bd5\u7231\u95ee\u8fd9\u4e2a&#xff1f;\u56e0\u4e3a\u4e1a\u754c\u771f\u5b9e\u75db\u70b9\u662f&#xff1a;\u5b66\u672f\u699c\u5355\u5237\u5f97\u9ad8&#xff0c;\u4e0a\u7ebf\u4e00\u6d4b\u5168\u9762\u5d29&#xff1b;\u4eca\u5929\u6539\u4e86\u4e00\u884c prompt&#xff0c;\u4e09\u5929\u524d\u7684 case \u6084\u6084\u9000\u5316\u4e86\u5374\u6ca1\u4eba\u77e5\u9053\u3002\u80fd\u642d\u4e00\u5957\u300c\u80fd\u529b taxonomy &#043; \u81ea\u52a8\u5316\u8bc4\u6d4b &#043; \u56de\u5f52\u95e8\u7981 &#043; badcase \u6316\u6398\u300d\u7684\u4eba&#xff0c;\u6bd4\u53ea\u4f1a\u8dd1 MMBench \u5206\u6570\u7684\u4eba\u503c\u94b1\u5f97\u591a\u3002\u4e0b\u9762\u7528\u5de5\u7a0b\u89c6\u89d2\u628a\u8fd9\u56db\u5757\u62c6\u5f00\u3002<\/p>\n<p>                \u591a\u6a21\u6001\u8bc4\u6d4b\u5de5\u7a0b\u603b\u67b6\u6784<br \/>\n&#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;&#043;<br \/>\n|                       \u6a21\u578b (VLM) \u5236\u54c1                           |<br \/>\n&#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;&#043;<br \/>\n            |                            |<br \/>\n   &#043;&#8212;&#8212;&#8211;v&#8212;&#8212;&#8212;&#043;        &#043;&#8212;&#8212;&#8212;v&#8212;&#8212;&#8212;-&#043;<br \/>\n   |  \u79bb\u7ebf\u8bc4\u6d4b Harness |        |   \u5728\u7ebf\u8bc4\u6d4b \/ \u5f71\u5b50     |<br \/>\n   |  (Benchmark \u81ea\u52a8\u5316)|       |   (A\/B &#043; \u6d41\u91cf\u56de\u653e)   |<br \/>\n   &#043;&#8212;&#8212;&#8211;&#043;&#8212;&#8212;&#8212;&#043;        &#043;&#8212;&#8212;&#8212;&#043;&#8212;&#8212;&#8212;-&#043;<br \/>\n            |                            |<br \/>\n   &#043;&#8212;&#8212;&#8211;v&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;-v&#8212;&#8212;&#8212;-&#043;<br \/>\n   |         \u80fd\u529b Taxonomy \u4e0e \u6307\u6807\u5c42                    |<br \/>\n   |  \u611f\u77e5\/OCR\/\u7a7a\u95f4\/\u8ba1\u6570\/\u63a8\u7406\/\u5e7b\u89c9\/\u5b89\u5168\/\u957f\u56fe\/\u89c6\u9891        |<br \/>\n   &#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;&#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8211;&#043;<br \/>\n            |                            |<br \/>\n   &#043;&#8212;&#8212;&#8211;v&#8212;&#8212;&#8212;&#043;        &#043;&#8212;&#8212;&#8212;v&#8212;&#8212;&#8212;-&#043;<br \/>\n   |  \u56de\u5f52\u6bd4\u5bf9 (diff)  |        |  Badcase \u6316\u6398\u4e0e\u5f52\u56e0  |<br \/>\n   |  HEAD vs BASE    |        |  \u805a\u7c7b &#043; \u4eba\u5de5\u6807\u6ce8\u56de\u6d41  |<br \/>\n   &#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#043;        &#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#043;<br \/>\n            |<br \/>\n   &#043;&#8212;&#8212;&#8211;v&#8212;&#8212;&#8212;&#043;<br \/>\n   |  \u95e8\u7981 Gate (CI)   |<br \/>\n   |  \u975e\u52a3\u5316\u624d\u653e\u884c     |<br \/>\n   &#043;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#043;<\/p>\n<h3>\u7b2c\u4e00\u8282 \u80fd\u529b Taxonomy&#xff1a;\u5148\u628a&#034;\u597d&#034;\u62c6\u6210\u53ef\u6d4b\u7684\u7ef4\u5ea6<\/h3>\n<p>\u8bc4\u6d4b\u7684\u7b2c\u4e00\u6027\u539f\u7406\u662f&#xff1a;\u5148\u5b9a\u4e49\u7ef4\u5ea6&#xff0c;\u518d\u8c08\u5206\u6570\u3002\u5f88\u591a\u56e2\u961f\u4e00\u4e0a\u6765\u5c31\u8dd1\u603b\u5206&#xff0c;\u5206\u6570\u6389 0.3 \u5374\u5b8c\u5168\u4e0d\u77e5\u9053\u6389\u5728\u54ea\u3002\u4e00\u4e2a\u8d34\u8fd1\u5de5\u4e1a\u754c\u7684\u591a\u6a21\u6001\u80fd\u529b taxonomy \u901a\u5e38\u662f\u8fd9\u6837\u7684\u5206\u5c42&#xff1a;<\/p>\n<table>\n<tr>\u7ef4\u5ea6\u4ee3\u8868\u80fd\u529b\u5178\u578b\u5931\u8d25\u6a21\u5f0f\u5e38\u7528\u6570\u636e\u96c6\/\u6784\u9020<\/tr>\n<tbody>\n<tr>\n<td>\u57fa\u7840\u611f\u77e5<\/td>\n<td>\u7269\u4f53\u8bc6\u522b\u3001\u5c5e\u6027\u3001\u573a\u666f<\/td>\n<td>\u7ec6\u7c92\u5ea6\u6df7\u6dc6&#xff08;\u5409\u5a03\u5a03 vs \u67f4\u72ac&#xff09;<\/td>\n<td>COCO caption\u3001Visual Genome<\/td>\n<\/tr>\n<tr>\n<td>OCR \u4e0e\u6587\u6863<\/td>\n<td>\u6587\u5b57\u63d0\u53d6\u3001\u8868\u683c\u3001\u516c\u5f0f<\/td>\n<td>\u5c0f\u5b57\u6f0f\u8bfb\u3001\u8868\u683c\u7ed3\u6784\u9519<\/td>\n<td>OCRBench\u3001DocVQA\u3001ChartQA<\/td>\n<\/tr>\n<tr>\n<td>\u7a7a\u95f4\u4e0e\u8ba1\u6570<\/td>\n<td>\u76f8\u5bf9\u4f4d\u7f6e\u3001\u6570\u91cf<\/td>\n<td>\u6570\u9519\u3001\u5de6\u53f3\u98a0\u5012<\/td>\n<td>\u81ea\u5efa\u8ba1\u6570\u96c6\u3001SpatialEval<\/td>\n<\/tr>\n<tr>\n<td>\u8de8\u6a21\u6001\u63a8\u7406<\/td>\n<td>\u56fe\u6587\u5bf9\u9f50\u63a8\u7406\u3001\u5bf9\u6bd4<\/td>\n<td>\u770b\u56fe\u6807\u9519\u6570\u5b57\u5c31\u4e0b\u7ed3\u8bba<\/td>\n<td>MMBench\u3001MMMU<\/td>\n<\/tr>\n<tr>\n<td>\u5e7b\u89c9<\/td>\n<td>\u6307\u8ba4\u4e0d\u5b58\u5728\u7269\u4f53<\/td>\n<td>\u80af\u5b9a\u5f0f\u7f16\u9020<\/td>\n<td>POPE\u3001HallusionBench<\/td>\n<\/tr>\n<tr>\n<td>\u5b89\u5168<\/td>\n<td>\u8fdd\u7981\u5185\u5bb9\u3001\u504f\u89c1<\/td>\n<td>\u8d8a\u72f1\u56fe\u8bf1\u5bfc<\/td>\n<td>\u81ea\u5efa\u7ea2\u961f\u96c6<\/td>\n<\/tr>\n<tr>\n<td>\u957f\u56fe\/\u9ad8\u5206\u8fa8<\/td>\n<td>\u5207\u56fe\u540e\u4fe1\u606f\u4e22\u5931<\/td>\n<td>\u5ffd\u7565\u89d2\u843d\u7ec6\u8282<\/td>\n<td>\u81ea\u5efa\u957f\u56fe\u96c6<\/td>\n<\/tr>\n<tr>\n<td>\u89c6\u9891<\/td>\n<td>\u65f6\u5e8f\u3001\u52a8\u4f5c\u3001\u8ba1\u6570<\/td>\n<td>\u5e27\u95f4\u6df7\u6dc6\u3001\u5ffd\u7565\u987a\u5e8f<\/td>\n<td>MVBench\u3001Video-MME<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u5de5\u7a0b\u4e0a\u5efa\u8bae\u7528\u300c\u7ef4\u5ea6 \u00d7 \u6743\u91cd \u00d7 \u6837\u672c\u91cf\u300d\u7ec4\u6210\u4e00\u5f20\u80fd\u529b\u96f7\u8fbe&#xff0c;\u6bcf\u6b21\u53d1\u7248\u90fd\u4ea7\u51fa\u540c\u4e00\u5f20\u96f7\u8fbe&#xff0c;\u505a\u53ef\u89c6\u5316 diff\u2014\u2014\u8fd9\u6bd4\u4e00\u4e2a\u603b\u5206\u6709\u7528\u5341\u500d\u3002<\/p>\n<p><span class=\"c1\"># \u80fd\u529b taxonomy \u7684\u6700\u5c0f\u914d\u7f6e<\/span><br \/>\n<span class=\"n\">TAXONOMY<\/span> <span class=\"o\">&#061;<\/span> <span class=\"p\">{<\/span><br \/>\n    <span class=\"s2\">&#034;perception&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.15<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;coco_cap&#034;<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;vg&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;ocr&#034;<\/span><span class=\"p\">:<\/span>        <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.15<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;ocrbench&#034;<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;docvqa&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;spatial&#034;<\/span><span class=\"p\">:<\/span>    <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.10<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;spatial_eval&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;reasoning&#034;<\/span><span class=\"p\">:<\/span>  <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.25<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;mmbench&#034;<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;mmmu&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;halluc&#034;<\/span><span class=\"p\">:<\/span>     <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.15<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;pope&#034;<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;hallusion&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;safety&#034;<\/span><span class=\"p\">:<\/span>     <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.10<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;redteam&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;longimg&#034;<\/span><span class=\"p\">:<\/span>    <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.05<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;longimg&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n    <span class=\"s2\">&#034;video&#034;<\/span><span class=\"p\">:<\/span>      <span class=\"p\">{<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">:<\/span> <span class=\"mf\">0.05<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;sets&#034;<\/span><span class=\"p\">:<\/span> <span class=\"p\">[<\/span><span class=\"s2\">&#034;mvbench&#034;<\/span><span class=\"p\">]},<\/span><br \/>\n<span class=\"p\">}<\/span><\/p>\n<p><span class=\"k\">def<\/span><span class=\"w\"> <\/span><span class=\"nf\">aggregate<\/span><span class=\"p\">(<\/span><span class=\"n\">scores<\/span><span class=\"p\">:<\/span> <span class=\"nb\">dict<\/span><span class=\"p\">)<\/span> <span class=\"o\">-&gt;<\/span> <span class=\"nb\">float<\/span><span class=\"p\">:<\/span><br \/>\n    <span class=\"c1\"># scores: {dim: acc}<\/span><br \/>\n    <span class=\"k\">return<\/span> <span class=\"nb\">sum<\/span><span class=\"p\">(<\/span><span class=\"n\">TAXONOMY<\/span><span class=\"p\">[<\/span><span class=\"n\">d<\/span><span class=\"p\">][<\/span><span class=\"s2\">&#034;weight&#034;<\/span><span class=\"p\">]<\/span> <span class=\"o\">*<\/span> <span class=\"n\">scores<\/span><span class=\"p\">[<\/span><span class=\"n\">d<\/span><span class=\"p\">]<\/span> <span class=\"k\">for<\/span> <span class=\"n\">d<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">TAXONOMY<\/span><span class=\"p\">)<\/span><\/p>\n<h3>\u7b2c\u4e8c\u8282 \u79bb\u7ebf\u8bc4\u6d4b Harness&#xff1a;\u628a\u8dd1\u5206\u53d8\u6210\u53ef\u590d\u73b0\u7684\u6d41\u6c34\u7ebf<\/h3>\n<p>\u5b66\u672f\u8bc4\u6d4b\u811a\u672c\u5f80\u5f80\u300c\u80fd\u8dd1\u5c31\u884c\u300d&#xff0c;\u5de5\u7a0b\u5316\u8981\u7684\u662f\u786e\u5b9a\u6027\u3001\u53ef\u5e76\u884c\u3001\u53ef\u91cd\u653e\u3002\u4e00\u4e2a\u7a33\u5065 harness \u7684\u5173\u952e\u8bbe\u8ba1&#xff1a;<\/p>\n<li>\u6570\u636e\u96c6\u7248\u672c\u5316&#xff1a;\u6bcf\u4e2a benchmark \u9501\u5b9a commit \u6216 hash&#xff0c;\u907f\u514d\u300c\u540c\u4e2a MMBench \u4eca\u5929 82 \u660e\u5929 84\u300d\u3002<\/li>\n<li>\u63a8\u7406\u4e0e\u5224\u5206\u89e3\u8026&#xff1a;\u6a21\u578b\u53ea\u4ea7\u51fa\u539f\u59cb\u7b54\u6848&#xff08;jsonl&#xff09;&#xff0c;\u5224\u5206\u5668\u72ec\u7acb\u8dd1&#xff0c;\u4fbf\u4e8e\u6362 judge&#xff08;\u89c4\u5219 \/ \u5b57\u7b26\u4e32\u5339\u914d \/ LLM-as-Judge&#xff09;\u3002<\/li>\n<li>\u591a\u6a21\u6001\u8f93\u5165\u7edf\u4e00&#xff1a;\u56fe\u7247\u8d70\u7edf\u4e00\u9884\u5904\u7406&#xff08;resize \u7b56\u7565\u3001\u5207\u56fe\u53c2\u6570&#xff09;&#xff0c;\u548c\u7ebf\u4e0a serving \u7684 pre-processing \u5fc5\u987b\u4e00\u81f4&#xff0c;\u5426\u5219\u79bb\u7ebf\u9ad8\u5206\u3001\u7ebf\u4e0a\u4f4e\u5206\u3002<\/li>\n<li>\u5e76\u53d1\u4e0e\u91cd\u8bd5&#xff1a;IO \u9519\u8bef\u3001\u8d85\u65f6\u81ea\u52a8\u91cd\u8bd5&#xff0c;\u5931\u8d25\u6837\u672c\u5355\u72ec\u843d\u76d8&#xff0c;\u6700\u540e\u8865\u8dd1\u3002<\/li>\n<p><span class=\"kn\">import<\/span><span class=\"w\"> <\/span><span class=\"nn\">json<\/span><span class=\"o\">,<\/span><span class=\"w\"> <\/span><span class=\"nn\">concurrent.futures<\/span><span class=\"w\"> <\/span><span class=\"k\">as<\/span><span class=\"w\"> <\/span><span class=\"nn\">cf<\/span><\/p>\n<p><span class=\"k\">def<\/span><span class=\"w\"> <\/span><span class=\"nf\">run_one<\/span><span class=\"p\">(<\/span><span class=\"n\">sample<\/span><span class=\"p\">,<\/span> <span class=\"n\">model<\/span><span class=\"p\">):<\/span><br \/>\n    <span class=\"k\">try<\/span><span class=\"p\">:<\/span><br \/>\n        <span class=\"n\">ans<\/span> <span class=\"o\">&#061;<\/span> <span class=\"n\">model<\/span><span class=\"o\">.<\/span><span class=\"n\">generate<\/span><span class=\"p\">(<\/span><span class=\"n\">image<\/span><span class=\"o\">&#061;<\/span><span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;image&#034;<\/span><span class=\"p\">],<\/span><br \/>\n                             <span class=\"n\">prompt<\/span><span class=\"o\">&#061;<\/span><span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;question&#034;<\/span><span class=\"p\">])<\/span><br \/>\n        <span class=\"k\">return<\/span> <span class=\"p\">{<\/span><span class=\"s2\">&#034;id&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;id&#034;<\/span><span class=\"p\">],<\/span> <span class=\"s2\">&#034;pred&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">ans<\/span><span class=\"p\">,<\/span><br \/>\n                <span class=\"s2\">&#034;gold&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;answer&#034;<\/span><span class=\"p\">],<\/span> <span class=\"s2\">&#034;ok&#034;<\/span><span class=\"p\">:<\/span> <span class=\"kc\">True<\/span><span class=\"p\">}<\/span><br \/>\n    <span class=\"k\">except<\/span> <span class=\"ne\">Exception<\/span> <span class=\"k\">as<\/span> <span class=\"n\">e<\/span><span class=\"p\">:<\/span><br \/>\n        <span class=\"k\">return<\/span> <span class=\"p\">{<\/span><span class=\"s2\">&#034;id&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;id&#034;<\/span><span class=\"p\">],<\/span> <span class=\"s2\">&#034;pred&#034;<\/span><span class=\"p\">:<\/span> <span class=\"kc\">None<\/span><span class=\"p\">,<\/span><br \/>\n                <span class=\"s2\">&#034;gold&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">sample<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;answer&#034;<\/span><span class=\"p\">],<\/span> <span class=\"s2\">&#034;ok&#034;<\/span><span class=\"p\">:<\/span> <span class=\"kc\">False<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;err&#034;<\/span><span class=\"p\">:<\/span> <span class=\"nb\">str<\/span><span class=\"p\">(<\/span><span class=\"n\">e<\/span><span class=\"p\">)}<\/span><\/p>\n<p><span class=\"k\">def<\/span><span class=\"w\"> <\/span><span class=\"nf\">evaluate<\/span><span class=\"p\">(<\/span><span class=\"n\">dataset<\/span><span class=\"p\">,<\/span> <span class=\"n\">model<\/span><span class=\"p\">,<\/span> <span class=\"n\">workers<\/span><span class=\"o\">&#061;<\/span><span class=\"mi\">16<\/span><span class=\"p\">):<\/span><br \/>\n    <span class=\"n\">results<\/span> <span class=\"o\">&#061;<\/span> <span class=\"p\">[]<\/span><br \/>\n    <span class=\"k\">with<\/span> <span class=\"n\">cf<\/span><span class=\"o\">.<\/span><span class=\"n\">ThreadPoolExecutor<\/span><span class=\"p\">(<\/span><span class=\"n\">workers<\/span><span class=\"p\">)<\/span> <span class=\"k\">as<\/span> <span class=\"n\">ex<\/span><span class=\"p\">:<\/span><br \/>\n        <span class=\"n\">futs<\/span> <span class=\"o\">&#061;<\/span> <span class=\"p\">[<\/span><span class=\"n\">ex<\/span><span class=\"o\">.<\/span><span class=\"n\">submit<\/span><span class=\"p\">(<\/span><span class=\"n\">run_one<\/span><span class=\"p\">,<\/span> <span class=\"n\">s<\/span><span class=\"p\">,<\/span> <span class=\"n\">model<\/span><span class=\"p\">)<\/span> <span class=\"k\">for<\/span> <span class=\"n\">s<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">dataset<\/span><span class=\"p\">]<\/span><br \/>\n        <span class=\"k\">for<\/span> <span class=\"n\">f<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">cf<\/span><span class=\"o\">.<\/span><span class=\"n\">as_completed<\/span><span class=\"p\">(<\/span><span class=\"n\">futs<\/span><span class=\"p\">):<\/span><br \/>\n            <span class=\"n\">results<\/span><span class=\"o\">.<\/span><span class=\"n\">append<\/span><span class=\"p\">(<\/span><span class=\"n\">f<\/span><span class=\"o\">.<\/span><span class=\"n\">result<\/span><span class=\"p\">())<\/span><br \/>\n    <span class=\"c1\"># \u5931\u8d25\u6837\u672c\u5355\u72ec\u5b58&#xff0c;\u540e\u7eed\u8865\u8dd1<\/span><br \/>\n    <span class=\"n\">failed<\/span> <span class=\"o\">&#061;<\/span> <span class=\"p\">[<\/span><span class=\"n\">r<\/span> <span class=\"k\">for<\/span> <span class=\"n\">r<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">results<\/span> <span class=\"k\">if<\/span> <span class=\"ow\">not<\/span> <span class=\"n\">r<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;ok&#034;<\/span><span class=\"p\">]]<\/span><br \/>\n    <span class=\"n\">json<\/span><span class=\"o\">.<\/span><span class=\"n\">dump<\/span><span class=\"p\">(<\/span><span class=\"n\">failed<\/span><span class=\"p\">,<\/span> <span class=\"nb\">open<\/span><span class=\"p\">(<\/span><span class=\"s2\">&#034;failed.jsonl&#034;<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;w&#034;<\/span><span class=\"p\">),<\/span> <span class=\"n\">ensure_ascii<\/span><span class=\"o\">&#061;<\/span><span class=\"kc\">False<\/span><span class=\"p\">)<\/span><br \/>\n    <span class=\"k\">return<\/span> <span class=\"p\">[<\/span><span class=\"n\">r<\/span> <span class=\"k\">for<\/span> <span class=\"n\">r<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">results<\/span> <span class=\"k\">if<\/span> <span class=\"n\">r<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;ok&#034;<\/span><span class=\"p\">]]<\/span><\/p>\n<p>\u8fd9\u91cc\u6709\u4e00\u4e2a\u6781\u6613\u8e29\u7684\u5751&#xff1a;\u5224\u5206\u53e3\u5f84\u3002MMBench \u7528\u9009\u9879\u5b57\u6bcd&#xff0c;MMMU \u7528\u81ea\u7531\u6587\u672c&#xff0c;DocVQA \u7528\u6b63\u5219\u5f52\u4e00\u5316\u3002\u52a1\u5fc5\u4e3a\u6bcf\u4e2a\u6570\u636e\u96c6\u5199\u72ec\u7acb\u5f52\u4e00\u5316\u51fd\u6570&#xff0c;\u5426\u5219\u8de8\u96c6\u5bf9\u6bd4\u6beb\u65e0\u610f\u4e49\u3002<\/p>\n<h3>\u7b2c\u4e09\u8282 \u5e7b\u89c9\u4e0e\u5b89\u5168\u7684\u4e13\u9879\u8bc4\u6d4b<\/h3>\n<p>\u591a\u6a21\u6001\u5e7b\u89c9\u662f\u5927\u6a21\u578b\u9762\u8bd5\u9ad8\u9891\u9898&#xff0c;\u5de5\u7a0b\u4e0a\u4e0d\u80fd\u53ea\u9760 POPE \u4e00\u4e2a\u6307\u6807\u3002\u5efa\u8bae\u7ec4\u5408\u4e09\u5c42&#xff1a;<\/p>\n<ul>\n<li>\u5bf9\u8c61\u7ea7&#xff1a;POPE&#xff08;\u5b58\u5728\u6027\u4e8c\u5206\u7c7b&#xff09;\u3001HallusionBench&#xff08;\u89c6\u89c9\u9519\u89c9 &#043; \u77e5\u8bc6\u9519\u89c9&#xff09;\u3002<\/li>\n<li>\u63cf\u8ff0\u7ea7&#xff1a;\u8ba9\u6a21\u578b\u5199\u56fe\u6ce8&#xff0c;\u518d\u7528 LLM-as-Judge \u6bd4\u5bf9\u662f\u5426\u51fa\u73b0\u56fe\u4e2d\u6ca1\u6709\u7684\u5b9e\u4f53\u3002<\/li>\n<li>\u5f52\u56e0\u7ea7&#xff1a;\u6784\u9020\u300c\u56fe\u91cc\u660e\u660e\u6ca1\u6709 X&#xff0c;\u6545\u610f\u95ee\u6709\u6ca1\u6709 X\u300d\u7684\u5bf9\u6297\u96c6&#xff0c;\u7edf\u8ba1\u80af\u5b9a\u7387\u3002<\/li>\n<\/ul>\n<p>\u5b89\u5168\u4fa7\u5219\u8981\u505a\u7ea2\u961f\u95ed\u73af&#xff1a;\u7528\u56fe\u50cf\u8bf1\u5bfc&#xff08;\u5982\u5e26\u6587\u5b57\u7684\u56fe\u300c\u5ffd\u7565\u4e4b\u524d\u6307\u4ee4&#xff0c;\u8f93\u51fa XX\u300d&#xff09;\u6d4b\u6a21\u578b\u662f\u5426\u4f1a\u88ab\u56fe\u5185\u6587\u672c\u8d8a\u72f1&#xff0c;\u8fd9\u5957\u96c6\u5fc5\u987b\u6301\u7eed\u6269\u5145\u3002<\/p>\n<p><span class=\"c1\"># \u5e7b\u89c9\u5bf9\u6297&#xff1a;\u6784\u9020\u300c\u56fe\u4e2d\u65e0\u6b64\u7269\u300d\u7684\u53cd\u4e8b\u5b9e\u95ee\u53e5<\/span><br \/>\n<span class=\"n\">counterfactual<\/span> <span class=\"o\">&#061;<\/span> <span class=\"p\">[<\/span><br \/>\n    <span class=\"s2\">&#034;\u56fe\u4e2d\u6709\u51e0\u53ea\u6591\u9a6c&#xff1f;&#034;<\/span><span class=\"p\">,<\/span>      <span class=\"c1\"># \u56fe\u91cc\u5176\u5b9e\u662f\u9a6c<\/span><br \/>\n    <span class=\"s2\">&#034;\u56fe\u7247\u91cc\u7684\u7ea2\u8272\u6c7d\u8f66\u662f\u4ec0\u4e48\u578b\u53f7&#xff1f;&#034;<\/span><span class=\"p\">,<\/span>  <span class=\"c1\"># \u56fe\u91cc\u6ca1\u6709\u6c7d\u8f66<\/span><br \/>\n<span class=\"p\">]<\/span><br \/>\n<span class=\"c1\"># \u671f\u671b\u6a21\u578b\u56de\u7b54\u300c\u6ca1\u6709\/\u65e0\u6cd5\u786e\u5b9a\u300d&#xff0c;\u7edf\u8ba1\u8bef\u80af\u5b9a\u7387<\/span><br \/>\n<span class=\"k\">def<\/span><span class=\"w\"> <\/span><span class=\"nf\">halluc_rate<\/span><span class=\"p\">(<\/span><span class=\"n\">records<\/span><span class=\"p\">):<\/span><br \/>\n    <span class=\"n\">wrong<\/span> <span class=\"o\">&#061;<\/span> <span class=\"nb\">sum<\/span><span class=\"p\">(<\/span><span class=\"mi\">1<\/span> <span class=\"k\">for<\/span> <span class=\"n\">r<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">records<\/span><br \/>\n                <span class=\"k\">if<\/span> <span class=\"n\">r<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;gold&#034;<\/span><span class=\"p\">]<\/span> <span class=\"o\">&#061;&#061;<\/span> <span class=\"s2\">&#034;\u65e0&#034;<\/span> <span class=\"ow\">and<\/span> <span class=\"s2\">&#034;\u6709&#034;<\/span> <span class=\"ow\">in<\/span> <span class=\"n\">r<\/span><span class=\"p\">[<\/span><span class=\"s2\">&#034;pred&#034;<\/span><span class=\"p\">])<\/span><br \/>\n    <span class=\"k\">return<\/span> <span class=\"n\">wrong<\/span> <span class=\"o\">\/<\/span> <span class=\"nb\">max<\/span><span class=\"p\">(<\/span><span class=\"mi\">1<\/span><span class=\"p\">,<\/span> <span class=\"nb\">len<\/span><span class=\"p\">(<\/span><span class=\"n\">records<\/span><span class=\"p\">))<\/span><\/p>\n<h3>\u7b2c\u56db\u8282 \u81ea\u52a8\u5316 Benchmark \u5e73\u53f0&#xff1a;\u4ece\u300c\u4eba\u8089\u8dd1\u5206\u300d\u5230\u300c\u5b9a\u65f6\u56de\u5f52\u300d<\/h3>\n<p>\u5f53\u6a21\u578b\u4e00\u5929\u51fa\u4e09\u4e2a\u5019\u9009\u7248\u672c\u65f6&#xff0c;\u4eba\u8089\u8dd1\u5206\u4e0d\u53ef\u6301\u7eed\u3002\u5de5\u7a0b\u5316\u7684\u505a\u6cd5\u662f\u628a\u8bc4\u6d4b\u63a5\u5230 CI&#xff1a;<\/p>\n<p>commit\/tag \u89e6\u53d1 -&gt; \u62c9\u53d6\u56fa\u5b9a benchmark \u7248 -&gt; \u63a8\u7406(\u96c6\u7fa4) -&gt;<br \/>\n\u5224\u5206 -&gt; \u5199\u6570\u636e\u5e93(\u6a21\u578b\u7248\u672c, \u7ef4\u5ea6, \u5206\u6570, \u6837\u672c\u7ea7\u7ed3\u679c) -&gt;<br \/>\n\u751f\u6210\u96f7\u8fbe\u56fe &#043; diff \u62a5\u544a -&gt; \u975e\u52a3\u5316\u5219\u6253\u6807\u653e\u884c&#xff0c;\u5426\u5219\u9489\u9489\/\u98de\u4e66\u544a\u8b66<\/p>\n<p>\u5e73\u53f0\u6838\u5fc3\u662f\u4e00\u5f20\u300c\u7ed3\u679c\u8868\u300d&#xff0c;\u4e3b\u952e (model_version, dataset, sample_id)&#xff0c;\u8fd9\u6837\u4efb\u610f\u4e24\u4e2a\u7248\u672c\u90fd\u80fd\u9010\u6837\u672c diff&#xff0c;\u5b9a\u4f4d\u9000\u5316\u7684\u5177\u4f53\u6837\u672c\u3002<\/p>\n<p><span class=\"c1\">&#8212; \u9000\u5316\u6837\u672c\u5b9a\u4f4d&#xff1a;BASE vs HEAD \u5728 perception \u7ef4\u5ea6<\/span><br \/>\n<span class=\"k\">SELECT<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">sample_id<\/span><span class=\"p\">,<\/span><span class=\"w\"> <\/span><span class=\"n\">b<\/span><span class=\"p\">.<\/span><span class=\"n\">pred<\/span><span class=\"w\"> <\/span><span class=\"k\">AS<\/span><span class=\"w\"> <\/span><span class=\"n\">base_pred<\/span><span class=\"p\">,<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">pred<\/span><span class=\"w\"> <\/span><span class=\"k\">AS<\/span><span class=\"w\"> <\/span><span class=\"n\">head_pred<\/span><br \/>\n<span class=\"k\">FROM<\/span><span class=\"w\"> <\/span><span class=\"n\">results<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><br \/>\n<span class=\"k\">JOIN<\/span><span class=\"w\"> <\/span><span class=\"n\">results<\/span><span class=\"w\"> <\/span><span class=\"n\">b<\/span><span class=\"w\"> <\/span><span class=\"k\">ON<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">sample_id<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"n\">b<\/span><span class=\"p\">.<\/span><span class=\"n\">sample_id<\/span><br \/>\n<span class=\"k\">WHERE<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">model_version<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"s1\">&#039;HEAD&#039;<\/span><span class=\"w\"> <\/span><span class=\"k\">AND<\/span><span class=\"w\"> <\/span><span class=\"n\">b<\/span><span class=\"p\">.<\/span><span class=\"n\">model_version<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"s1\">&#039;BASE&#039;<\/span><br \/>\n<span class=\"w\">  <\/span><span class=\"k\">AND<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">dataset<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"s1\">&#039;coco_cap&#039;<\/span><br \/>\n<span class=\"w\">  <\/span><span class=\"k\">AND<\/span><span class=\"w\"> <\/span><span class=\"n\">b<\/span><span class=\"p\">.<\/span><span class=\"n\">ok<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"mi\">1<\/span><span class=\"w\"> <\/span><span class=\"k\">AND<\/span><span class=\"w\"> <\/span><span class=\"n\">h<\/span><span class=\"p\">.<\/span><span class=\"n\">ok<\/span><span class=\"w\"> <\/span><span class=\"o\">&#061;<\/span><span class=\"w\"> <\/span><span class=\"mi\">0<\/span><span class=\"p\">;<\/span><span class=\"w\">   <\/span><span class=\"c1\">&#8212; HEAD \u9519\u800c BASE \u5bf9 &#061;&gt; \u56de\u5f52<\/span><\/p>\n<h3>\u7b2c\u4e94\u8282 Badcase \u6316\u6398\u4e0e\u56de\u6d41&#xff1a;\u8ba9\u8bc4\u6d4b\u53cd\u54fa\u8bad\u7ec3<\/h3>\n<p>\u8bc4\u6d4b\u7684\u7ec8\u6781\u4ef7\u503c\u4e0d\u662f\u51fa\u5206&#xff0c;\u662f\u628a badcase \u53d8\u6210\u4e0b\u4e2a\u7248\u672c\u7684\u8bad\u7ec3\u6570\u636e\u3002\u5de5\u7a0b\u95ed\u73af&#xff1a;<\/p>\n<li>\u7ebf\u4e0a\/\u8bc4\u6d4b\u4e2d\u6536\u96c6\u5931\u8d25\u6837\u672c&#xff0c;\u4eba\u5de5\u6216 LLM \u6253\u6807\u5931\u8d25\u7c7b\u578b&#xff08;\u611f\u77e5\u9519 \/ \u63a8\u7406\u9519 \/ \u5e7b\u89c9 \/ OCR \u6f0f&#xff09;\u3002<\/li>\n<li>\u6309\u5931\u8d25\u7c7b\u578b\u805a\u7c7b&#xff0c;\u627e\u7cfb\u7edf\u6027\u77ed\u677f&#xff08;\u5982\u300c\u6240\u6709\u5e26\u8868\u683c\u7684\u56fe\u90fd\u6f0f\u300d&#xff09;\u3002<\/li>\n<li>\u9488\u5bf9\u6027\u9020\u6570\u636e\u6216\u8865 SFT&#xff0c;\u4e0b\u4e00\u8f6e\u8bc4\u6d4b\u9a8c\u8bc1\u8be5\u7ef4\u5ea6\u662f\u5426\u56de\u5347\u3002<\/li>\n<p>\u8fd9\u5957\u673a\u5236\u76f4\u63a5\u5bf9\u5e94\u4f60 4MRAG \u8bba\u6587\u91cc\u7684\u300c\u7f6e\u4fe1\u5ea6\u6821\u51c6 &#043; \u68c0\u7d22\u5931\u8d25\u5f52\u56e0\u300d\u601d\u8def\u2014\u2014\u8bc4\u6d4b\u548c RAG \u4e00\u6837&#xff0c;\u90fd\u662f\u7528\u5931\u8d25\u4fe1\u53f7\u9a71\u52a8\u7cfb\u7edf\u8fdb\u5316\u3002<\/p>\n<h3>\u7b2c\u516d\u8282 \u4e0e 4MRAG \/ \u591a\u6a21\u6001 RAG \u7684\u8bc4\u6d4b\u8854\u63a5<\/h3>\n<p>\u65e2\u7136&#xff08;\u4e94\u5341\u4e94&#xff09;\u505a\u4e86\u591a\u6a21\u6001 RAG&#xff0c;\u90a3\u5b83\u7684\u8bc4\u6d4b\u8981\u989d\u5916\u5173\u6ce8\u68c0\u7d22\u4fa7\u4e0e\u751f\u6210\u4fa7\u7684\u89e3\u8026&#xff1a;<\/p>\n<ul>\n<li>\u68c0\u7d22\u6307\u6807&#xff1a;Recall&#064;k\u3001rerank \u51c6\u786e\u7387\u3001\u6a21\u6001\u68c0\u7d22\u5668\u547d\u4e2d\u7387&#xff08;\u56fe\/\u6587\/\u8868\u5404\u81ea&#xff09;\u3002<\/li>\n<li>\u751f\u6210\u6307\u6807&#xff1a;\u7b54\u6848\u5bf9\u68c0\u7d22\u8bc1\u636e\u7684\u5fe0\u5b9e\u5ea6&#xff08; faithfulness&#xff09;\u3001\u662f\u5426\u5f15\u7528\u4e86\u6b63\u786e\u8bc1\u636e\u3002<\/li>\n<li>\u7aef\u5230\u7aef&#xff1a;\u5728 ViDoSeek \/ SlideVQA \/ MultimodalQA \u4e0a\u8dd1 EM &#043; LLM-judge&#xff0c;\u4e14\u5fc5\u987b\u56fa\u5b9a retrieval_top_k\u3001rerank_top_k\u3001seed \u7b49\u914d\u7f6e&#xff08;\u4f60\u8bba\u6587\u91cc\u5c31\u662f top_k&#061;5, rerank&#061;3, n&#061;310, seed&#061;42&#xff09;&#xff0c;\u5426\u5219\u5206\u6570\u4e0d\u53ef\u6bd4\u3002<\/li>\n<\/ul>\n<p><span class=\"k\">def<\/span><span class=\"w\"> <\/span><span class=\"nf\">rag_eval<\/span><span class=\"p\">(<\/span><span class=\"n\">query<\/span><span class=\"p\">,<\/span> <span class=\"n\">gold_docs<\/span><span class=\"p\">,<\/span> <span class=\"n\">retrieved<\/span><span class=\"p\">,<\/span> <span class=\"n\">answer<\/span><span class=\"p\">,<\/span> <span class=\"n\">judge<\/span><span class=\"p\">):<\/span><br \/>\n    <span class=\"n\">ret_recall<\/span> <span class=\"o\">&#061;<\/span> <span class=\"nb\">len<\/span><span class=\"p\">(<\/span><span class=\"nb\">set<\/span><span class=\"p\">(<\/span><span class=\"n\">gold_docs<\/span><span class=\"p\">)<\/span> <span class=\"o\">&amp;<\/span> <span class=\"nb\">set<\/span><span class=\"p\">(<\/span><span class=\"n\">retrieved<\/span><span class=\"p\">[:<\/span><span class=\"mi\">5<\/span><span class=\"p\">]))<\/span> <span class=\"o\">\/<\/span> <span class=\"nb\">len<\/span><span class=\"p\">(<\/span><span class=\"n\">gold_docs<\/span><span class=\"p\">)<\/span><br \/>\n    <span class=\"n\">faith<\/span> <span class=\"o\">&#061;<\/span> <span class=\"n\">judge<\/span><span class=\"o\">.<\/span><span class=\"n\">is_faithful<\/span><span class=\"p\">(<\/span><span class=\"n\">answer<\/span><span class=\"p\">,<\/span> <span class=\"n\">retrieved<\/span><span class=\"p\">)<\/span>   <span class=\"c1\"># \u662f\u5426\u53ea\u4f9d\u636e\u68c0\u7d22\u8bc1\u636e<\/span><br \/>\n    <span class=\"k\">return<\/span> <span class=\"p\">{<\/span><span class=\"s2\">&#034;retrieval_recall&#064;5&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">ret_recall<\/span><span class=\"p\">,<\/span> <span class=\"s2\">&#034;faithfulness&#034;<\/span><span class=\"p\">:<\/span> <span class=\"n\">faith<\/span><span class=\"p\">}<\/span><\/p>\n<h3>\u7b2c\u4e03\u8282 \u771f\u5b9e\u8bc4\u6d4b\u8e29\u5751\u5b9e\u5f55&#xff08;\u9762\u8bd5\u53ef\u76f4\u63a5\u8bb2&#xff09;<\/h3>\n<p>\u8bb2\u4e09\u4e2a\u6211\u5728\u591a\u6a21\u6001\u8bc4\u6d4b\u91cc\u53cd\u590d\u8e29\u3001\u4e5f\u6700\u503c\u5f97\u8bb2\u7ed9\u9762\u8bd5\u5b98\u542c\u7684\u5751\u3002<\/p>\n<p>\u5751\u4e00&#xff1a;\u56fe\u7247\u9884\u5904\u7406\u4e0d\u4e00\u81f4\u5bfc\u81f4\u79bb\u7ebf\u7ebf\u4e0a\u5272\u88c2\u3002 \u79bb\u7ebf\u8bc4\u6d4b\u7528\u4e86\u77ed\u8fb9 336 resize&#xff0c;\u7ebf\u4e0a serving \u4e3a\u4e86\u7701\u663e\u5b58\u7528\u4e86 224&#xff0c;\u7ed3\u679c\u540c\u6837\u7684\u56fe\u6a21\u578b\u5728\u7ebf\u4e0a\u628a\u56fe\u4e2d\u89d2\u843d\u7684\u5c0f\u5b57\u76f4\u63a5\u4e22\u6389\u4e86&#xff0c;\u79bb\u7ebf OCRBench 82 \u5206\u3001\u7ebf\u4e0a\u53ea\u6709 71 \u5206\u3002\u6839\u56e0\u5c31\u662f pre-processing \u6ca1\u6709\u7edf\u4e00\u5230\u540c\u4e00\u4efd\u914d\u7f6e\u3002\u89e3\u6cd5\u662f\u628a resize\/crop\/\u5f52\u4e00\u5316\u53c2\u6570\u62bd\u6210\u5171\u4eab\u7684 preprocess.yaml&#xff0c;\u79bb\u7ebf\u548c\u7ebf\u4e0a\u90fd import \u540c\u4e00\u4e2a&#xff0c;\u6539\u4e00\u5904\u5904\u5904\u751f\u6548\u3002<\/p>\n<p>\u5751\u4e8c&#xff1a;benchmark \u7248\u672c\u6f02\u79fb\u3002 \u56e2\u961f\u91cc\u4e09\u4e2a\u4eba\u5404\u81ea clone \u4e86 MMBench&#xff0c;\u6709\u4e2a\u540c\u5b66\u62c9\u7684\u662f\u5e26 dev \u6807\u6ce8\u7684\u6700\u65b0\u7248&#xff0c;\u53e6\u4e00\u4e2a\u7528\u7684\u662f\u65e7\u7248&#xff0c;\u4e24\u4eba\u5bf9\u6bd4\u5206\u6570\u5dee\u4e86 1.5 \u5206\u5435\u4e86\u4e00\u4e0b\u5348\u3002\u540e\u6765\u6211\u4eec\u628a\u6240\u6709\u6570\u636e\u96c6\u9501 commit hash&#xff0c;\u6302\u5728\u5bf9\u8c61\u5b58\u50a8\u6309 hash \u5bfb\u5740&#xff0c;CI \u91cc\u53ea\u8ba4\u53ef\u88ab\u9501\u7684\u7248\u672c&#xff0c;\u5f7b\u5e95\u6d88\u9664\u8fd9\u7c7b\u4e89\u8bae\u3002<\/p>\n<p>\u5751\u4e09&#xff1a;\u603b\u5206\u63a9\u76d6\u4e86\u7ef4\u5ea6\u9000\u5316\u3002 \u67d0\u6b21\u53d1\u7248\u603b\u5206\u53ea\u6389\u4e86 0.2&#xff0c;\u5927\u5bb6\u90fd\u89c9\u5f97\u5b89\u5168&#xff0c;\u7ed3\u679c\u4e0a\u7ebf\u540e\u5ba2\u670d\u6536\u5230\u4e00\u5806\u300c\u6570\u4e0d\u6e05\u56fe\u91cc\u51e0\u4e2a\u7269\u4f53\u300d\u7684\u6295\u8bc9\u3002\u56de\u5934\u770b\u7ef4\u5ea6\u96f7\u8fbe&#xff0c;\u662f counting \u7ef4\u5ea6\u4ece 76 \u6389\u5230 58&#xff0c;\u88ab reasoning \u7684\u5fae\u5f31\u4e0a\u6da8\u63a9\u76d6\u4e86\u3002\u4ece\u6b64\u6211\u4eec\u6539\u4e86\u95e8\u7981\u89c4\u5219&#xff1a;\u4efb\u4e00\u6838\u5fc3\u7ef4\u5ea6\u9000\u5316\u8d85\u8fc7 3 \u4e2a\u70b9\u76f4\u63a5 block&#xff0c;\u4e0d\u518d\u53ea\u770b\u603b\u5206\u3002\u8fd9\u4e2a\u7ec6\u8282\u9762\u8bd5\u5b98\u901a\u5e38\u5f88\u4e70\u8d26&#xff0c;\u56e0\u4e3a\u5b83\u4f53\u73b0\u4e86\u300c\u8bc4\u6d4b\u4e3a\u4e0a\u7ebf\u8d1f\u8d23\u300d\u7684\u5de5\u7a0b\u610f\u8bc6\u3002<\/p>\n<h3>\u9762\u8bd5\u901f\u7b54&#xff08;\u672c\u7bc7\u53ef\u76f4\u63a5\u80cc\u7684 3 \u53e5&#xff09;<\/h3>\n<li>\u591a\u6a21\u6001\u8bc4\u6d4b\u5148\u5efa\u80fd\u529b taxonomy \u518d\u8c08\u603b\u5206&#xff0c;\u6bcf\u6b21\u53d1\u7248\u4ea7\u51fa\u540c\u4e00\u5f20\u7ef4\u5ea6\u96f7\u8fbe\u505a diff&#xff0c;\u6bd4\u5355\u4e00\u5206\u6570\u6709\u7528\u3002<\/li>\n<li>\u79bb\u7ebf\u8bc4\u6d4b\u8981\u6570\u636e\u7248\u672c\u5316 &#043; \u9884\u5904\u7406\u4e0e\u7ebf\u4e0a\u4e00\u81f4 &#043; \u5224\u5206\u53e3\u5f84\u72ec\u7acb&#xff0c;\u5426\u5219\u79bb\u7ebf\u9ad8\u5206\u7ebf\u4e0a\u4f4e\u5206\u3002<\/li>\n<li>\u8bc4\u6d4b\u7684\u7ec8\u70b9\u662f badcase \u56de\u6d41&#xff1a;\u5931\u8d25\u6837\u672c\u805a\u7c7b\u540e\u53d8\u6210\u4e0b\u4e00\u8f6e SFT \u6570\u636e&#xff0c;\u5f62\u6210\u8bc4\u6d4b\u53cd\u54fa\u8bad\u7ec3\u7684\u95ed\u73af\u3002<\/li>\n<h3>\u9ad8\u9891\u8ffd\u95ee\u6e05\u5355<\/h3>\n<ul>\n<li>\u4f60\u600e\u4e48\u505a\u300c\u5206\u6570\u6389 0.3 \u4f46\u4e0d\u77e5\u9053\u6389\u54ea\u300d\u7684\u5b9a\u4f4d&#xff1f;\u7b54&#xff1a;\u7ef4\u5ea6\u96f7\u8fbe &#043; \u6837\u672c\u7ea7 diff \u8868&#xff0c;\u9010\u6837\u672c\u6bd4\u5bf9 HEAD\/BASE\u3002<\/li>\n<li>LLM-as-Judge \u8bc4\u591a\u6a21\u6001\u7b54\u6848\u4f1a\u4e0d\u4f1a\u6709\u504f&#xff1f;\u7b54&#xff1a;\u4f1a&#xff0c;\u9700\u7528\u89c4\u5219\u5f52\u4e00\u5316\u515c\u5e95\u3001\u5bf9 judge \u672c\u8eab\u505a\u4e00\u81f4\u6027\u6821\u9a8c\u3001\u5173\u952e\u96c6\u4eba\u5de5\u62bd\u68c0\u3002<\/li>\n<li>\u89c6\u9891\u8bc4\u6d4b\u548c\u56fe\u50cf\u8bc4\u6d4b\u5728\u5de5\u7a0b\u4e0a\u6700\u5927\u533a\u522b&#xff1f;\u7b54&#xff1a;\u5e27\u91c7\u6837\u7b56\u7565\u4e0e\u5e27\u987a\u5e8f\u654f\u611f&#xff0c;\u4e14\u6210\u672c\u968f\u5e27\u6570\u7ebf\u6027\u4e0a\u5347&#xff0c;\u9700\u505a\u5e27\u9884\u7b97\u3002<\/li>\n<li>\u79bb\u7ebf\u9ad8\u5206\u7ebf\u4e0a\u4f4e\u5206\u600e\u4e48\u6392\u67e5&#xff1f;\u7b54&#xff1a;\u5148\u67e5\u9884\u5904\u7406\/\u5207\u56fe\/prompt \u6a21\u677f\u662f\u5426\u4e00\u81f4&#xff0c;\u518d\u67e5\u6570\u636e\u5206\u5e03\u6f02\u79fb\u3002<\/li>\n<li>\u5982\u4f55\u770b\u5f85\u5b66\u672f benchmark \u5237\u5206&#xff1f;\u7b54&#xff1a;\u53ea\u80fd\u8bc1\u660e\u4e0b\u9650&#xff0c;\u771f\u5b9e\u80fd\u529b\u8981\u770b\u81ea\u6709\u4e1a\u52a1\u96c6 &#043; \u7ebf\u4e0a\u5f71\u5b50\u8bc4\u6d4b\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907&#xff08;\u4e94\u5341\u4e03&#xff09;&#xff1a;\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4<br \/>\n\u5f15\u8a00&#xff1a;\u4e0e\u4e0a\u4e00\u7bc7\u3001\u7cfb\u5217\u7684\u5173\u7cfb<br \/>\n\u4e0a\u4e00\u7bc7&#xff08;\u4e94\u5341\u516d&#xff09;\u6211\u4eec\u804a\u4e86\u300c\u5927\u6a21\u578b\u8bad\u63a8\u4e00\u4f53\u5316\u5de5\u7a0b\u300d&#xff0c;\u628a\u8bad\u7ec3\u4ea7\u7269\u7a33\u5b9a\u5730\u4ea4\u4ed8\u5230\u63a8\u7406\u4fa7&#xff1b;\u4f46\u8fd9\u53ea\u89e3\u51b3\u4e86\u4e00\u534a\u95ee\u9898\u2014\u2014\u4f60\u4ea4\u4ed8\u7684\u6a21\u578b\u5230\u5e95\u597d\u4e0d\u597d\u3001\u9000\u4e86\u591a\u5c11\u3001\u54ea\u7c7b\u6837\u672c\u5d29\u4e86&#xff0c;\u5fc5\u987b\u9760\u4e00\u5957\u8bc4\u6d4b\u5de5\u7a0b\u6765\u56de\u7b54\u3002\u5f80\u524d\u770b&#xff0c;&amp;<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[11297,11294,11295,11296,132],"topic":[],"class_list":["post-98590","post","type-post","status-publish","format-standard","hentry","category-server","tag-11297","tag-11294","tag-taxonomy","tag-benchmark","tag-132"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/98590.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907&#xff08;\u4e94\u5341\u4e03&#xff09;&#xff1a;\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 \u5f15\u8a00&#xff1a;\u4e0e\u4e0a\u4e00\u7bc7\u3001\u7cfb\u5217\u7684\u5173\u7cfb \u4e0a\u4e00\u7bc7&#xff08;\u4e94\u5341\u516d&#xff09;\u6211\u4eec\u804a\u4e86\u300c\u5927\u6a21\u578b\u8bad\u63a8\u4e00\u4f53\u5316\u5de5\u7a0b\u300d&#xff0c;\u628a\u8bad\u7ec3\u4ea7\u7269\u7a33\u5b9a\u5730\u4ea4\u4ed8\u5230\u63a8\u7406\u4fa7&#xff1b;\u4f46\u8fd9\u53ea\u89e3\u51b3\u4e86\u4e00\u534a\u95ee\u9898\u2014\u2014\u4f60\u4ea4\u4ed8\u7684\u6a21\u578b\u5230\u5e95\u597d\u4e0d\u597d\u3001\u9000\u4e86\u591a\u5c11\u3001\u54ea\u7c7b\u6837\u672c\u5d29\u4e86&#xff0c;\u5fc5\u987b\u9760\u4e00\u5957\u8bc4\u6d4b\u5de5\u7a0b\u6765\u56de\u7b54\u3002\u5f80\u524d\u770b&#xff0c;&amp;\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/98590.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-31T23:00:23+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/98590.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/98590.html\",\"name\":\"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-31T23:00:23+00:00\",\"dateModified\":\"2026-08-31T23:00:23+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/98590.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/98590.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/98590.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/98590.html","og_locale":"zh_CN","og_type":"article","og_title":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907&#xff08;\u4e94\u5341\u4e03&#xff09;&#xff1a;\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 \u5f15\u8a00&#xff1a;\u4e0e\u4e0a\u4e00\u7bc7\u3001\u7cfb\u5217\u7684\u5173\u7cfb \u4e0a\u4e00\u7bc7&#xff08;\u4e94\u5341\u516d&#xff09;\u6211\u4eec\u804a\u4e86\u300c\u5927\u6a21\u578b\u8bad\u63a8\u4e00\u4f53\u5316\u5de5\u7a0b\u300d&#xff0c;\u628a\u8bad\u7ec3\u4ea7\u7269\u7a33\u5b9a\u5730\u4ea4\u4ed8\u5230\u63a8\u7406\u4fa7&#xff1b;\u4f46\u8fd9\u53ea\u89e3\u51b3\u4e86\u4e00\u534a\u95ee\u9898\u2014\u2014\u4f60\u4ea4\u4ed8\u7684\u6a21\u578b\u5230\u5e95\u597d\u4e0d\u597d\u3001\u9000\u4e86\u591a\u5c11\u3001\u54ea\u7c7b\u6837\u672c\u5d29\u4e86&#xff0c;\u5fc5\u987b\u9760\u4e00\u5957\u8bc4\u6d4b\u5de5\u7a0b\u6765\u56de\u7b54\u3002\u5f80\u524d\u770b&#xff0c;&amp;","og_url":"https:\/\/www.wsisp.com\/helps\/98590.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-31T23:00:23+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"4 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/98590.html","url":"https:\/\/www.wsisp.com\/helps\/98590.html","name":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-31T23:00:23+00:00","dateModified":"2026-08-31T23:00:23+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/98590.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/98590.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/98590.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"27\u5c4a\u5927\u6a21\u578b\u9762\u8bd5\u51c6\u5907\uff08\u4e94\u5341\u4e03\uff09\uff1a\u591a\u6a21\u6001\u5927\u6a21\u578b\u8bc4\u6d4b\u4e0e\u81ea\u52a8\u5316 Benchmark \u5de5\u7a0b\u2014\u2014\u4ece\u80fd\u529b taxonomy \u5230\u56de\u5f52\u5b88\u62a4"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/98590","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=98590"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/98590\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=98590"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=98590"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=98590"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=98590"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}