{"id":848,"date":"2026-07-16T23:18:23","date_gmt":"2026-07-16T15:18:23","guid":{"rendered":"https:\/\/blog.itsmygo.uk\/2026\/07\/16\/%e3%80%90mlflow-llms-agents-%e5%af%a6%e6%88%b0%e3%80%9102-%e7%94%a8-llm-as-a-judge-%e8%a9%95%e4%bc%b0%e4%bd%a0%e7%9a%84%e6%a8%a1%e5%9e%8b%e8%bc%b8%e5%87%ba\/"},"modified":"2026-07-16T23:18:23","modified_gmt":"2026-07-16T15:18:23","slug":"%e3%80%90mlflow-llms-agents-%e5%af%a6%e6%88%b0%e3%80%9102-%e7%94%a8-llm-as-a-judge-%e8%a9%95%e4%bc%b0%e4%bd%a0%e7%9a%84%e6%a8%a1%e5%9e%8b%e8%bc%b8%e5%87%ba","status":"publish","type":"post","link":"https:\/\/blog.itsmygo.uk\/wordpress\/2026\/07\/16\/%e3%80%90mlflow-llms-agents-%e5%af%a6%e6%88%b0%e3%80%9102-%e7%94%a8-llm-as-a-judge-%e8%a9%95%e4%bc%b0%e4%bd%a0%e7%9a%84%e6%a8%a1%e5%9e%8b%e8%bc%b8%e5%87%ba\/","title":{"rendered":"\u3010MLflow LLMs &#038; Agents \u5be6\u6230\u3011#02 \u7528 LLM-as-a-Judge \u8a55\u4f30\u4f60\u7684\u6a21\u578b\u8f38\u51fa"},"content":{"rendered":"\n<div id=\"quiz-836595d3\">\n<style>\n#quiz-836595d3 .question {\n        background: white; border-radius: 8px; padding: 20px; margin: 20px 0;\n        box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);\n      }#quiz-836595d3 .question h3 { margin-top: 0; color: #333; }#quiz-836595d3 .options { list-style: none; padding: 0; }#quiz-836595d3 .options li {\n        padding: 12px 15px; margin: 8px 0; border: 2px solid #ddd;\n        border-radius: 6px; cursor: pointer; transition: all 0.2s;\n      }#quiz-836595d3 .options li:hover:not(.disabled) { border-color: #4caf50; background: #f0fff0; }#quiz-836595d3 .options li.correct { border-color: #4caf50; background: #e8f5e9; }#quiz-836595d3 .options li.wrong { border-color: #f44336; background: #ffebee; }#quiz-836595d3 .options li.disabled { cursor: default; opacity: 0.7; }#quiz-836595d3 .options li.show-answer { border-color: #4caf50; background: #e8f5e9; }#quiz-836595d3 .feedback { margin-top: 15px; padding: 12px; border-radius: 6px; display: none; }#quiz-836595d3 .feedback.correct { background: #e8f5e9; color: #2e7d32; display: block; }#quiz-836595d3 .feedback.wrong { background: #ffebee; color: #c62828; display: block; }#quiz-836595d3 .score {\n        text-align: center; font-size: 1.5em; padding: 20px; background: white;\n        border-radius: 8px; margin-top: 20px;\n      }#quiz-836595d3 .code-block {\n        background: #1e1e1e; color: #d4d4d4; padding: 14px; border-radius: 6px;\n        font-family: \"Monaco\", \"Consolas\", \"Courier New\", monospace; font-size: 0.9em;\n        overflow-x: auto; margin: 12px 0; white-space: pre-wrap;\n      }\n<\/style>\n<h1>\u6e2c\u9a57\uff1a\u7528 LLM-as-a-Judge \u8a55\u4f30\u4f60\u7684\u6a21\u578b\u8f38\u51fa<\/h1>\n    <p>\u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50\u679c<\/p>\n\n    <div class=\"question\" id=\"q1\">\n      <h3>1. \u6587\u7ae0\u6307\u51fa\uff0c\u70ba\u4ec0\u9ebc LLM \u7684\u8f38\u51fa\u300c\u96e3\u4ee5\u7528\u50b3\u7d71 metric\uff08\u5982 exact match\u3001BLEU\uff09\u8a55\u4f30\u300d\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(1, 'A', 'B')\">A. \u56e0\u70ba LLM \u7684\u8f38\u51fa\u592a\u9577\uff0c\u7121\u6cd5\u8a08\u7b97<\/li>\n        <li onclick=\"checkAnswer(1, 'B', 'B')\">B. \u56e0\u70ba\u540c\u4e00\u500b\u554f\u984c\u7684\u597d\u7b54\u6848\u6709\u7121\u9650\u591a\u7a2e\uff0c\u5b57\u4e32\u6bd4\u5c0d\u6703\u628a\u597d\u7b54\u6848\u5224\u6210\u932f<\/li>\n        <li onclick=\"checkAnswer(1, 'C', 'B')\">C. \u56e0\u70ba\u50b3\u7d71 metric \u53ea\u80fd\u7528\u5728\u5716\u7247\u4e0a<\/li>\n        <li onclick=\"checkAnswer(1, 'D', 'B')\">D. \u56e0\u70ba LLM \u4e0d\u6703\u8f38\u51fa\u6587\u5b57\uff0c\u53ea\u6703\u8f38\u51fa\u5411\u91cf<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb1\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q2\">\n      <h3>2. \u5728\u8a55\u4f30\u8cc7\u6599\u96c6\u4e2d\uff0c\u4e00\u7b46\u8cc7\u6599\u7684 <code>inputs<\/code> \u548c <code>expectations<\/code> \u5206\u5225\u4ee3\u8868\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(2, 'A', 'C')\">A. inputs \u662f\u8a55\u5be9\u6a21\u578b\u540d\u7a31\uff1bexpectations \u662f\u5206\u6578\u9580\u6abb<\/li>\n        <li onclick=\"checkAnswer(2, 'B', 'C')\">B. inputs \u662f\u6a19\u6e96\u7b54\u6848\uff1bexpectations \u662f\u984c\u76ee<\/li>\n        <li onclick=\"checkAnswer(2, 'C', 'C')\">C. inputs \u662f\u9935\u7d66 app \u7684\u8f38\u5165\uff08\u984c\u76ee\/context\uff09\uff1bexpectations \u662f\u4f60\u5c0d\u9019\u984c\u7684\u671f\u671b\uff08\u6a19\u6e96\u7b54\u6848\u7b49\uff09<\/li>\n        <li onclick=\"checkAnswer(2, 'D', 'C')\">D. \u5169\u8005\u90fd\u662f\u8a55\u5be9\u6253\u5b8c\u5206\u5f8c\u624d\u7522\u751f\u7684\u7d50\u679c<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb2\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q3\">\n      <h3>3. \u95dc\u65bc\u5167\u5efa scorer \u9700\u4e0d\u9700\u8981\u300c\u6a19\u6e96\u7b54\u6848\uff08expected_response\uff09\u300d\uff0c\u4e0b\u5217\u4f55\u8005\u6b63\u78ba\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(3, 'A', 'A')\">A. <code>Correctness<\/code> \u9700\u8981\u6a19\u6e96\u7b54\u6848\uff1b<code>RelevanceToQuery<\/code> \u4e0d\u9700\u8981<\/li>\n        <li onclick=\"checkAnswer(3, 'B', 'A')\">B. \u6240\u6709 scorer \u90fd\u4e00\u5b9a\u8981\u6a19\u6e96\u7b54\u6848\u624d\u80fd\u8dd1<\/li>\n        <li onclick=\"checkAnswer(3, 'C', 'A')\">C. \u6240\u6709 scorer \u90fd\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848<\/li>\n        <li onclick=\"checkAnswer(3, 'D', 'A')\">D. <code>RelevanceToQuery<\/code> \u9700\u8981\u6a19\u6e96\u7b54\u6848\uff1b<code>Correctness<\/code> \u4e0d\u9700\u8981<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb3\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q4\">\n      <h3>4. \u5728 <code>make_judge<\/code> \u5efa\u7acb\u81ea\u8a02 judge \u6642\uff0c\u5224\u6e96\uff08rubric\uff09\u5177\u9ad4\u662f\u5beb\u5728\u54ea\u500b\u90e8\u5206\uff1f<\/h3>\n      <div class=\"code-block\">judge = make_judge(\n    name=&#8221;conciseness&#8221;,\n    instructions=&#8221;&#8230;{{ inputs }}&#8230;{{ outputs }}&#8230;pass\/fail&#8230;&#8221;,\n    model=&#8221;openai:\/gpt-4o&#8221;,\n)<\/div>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(4, 'A', 'C')\">A. \u5beb\u5728 <code>name<\/code> \u9019\u500b\u540d\u5b57\u88e1<\/li>\n        <li onclick=\"checkAnswer(4, 'B', 'C')\">B. \u5beb\u5728 <code>model<\/code> \u53c3\u6578\u88e1<\/li>\n        <li onclick=\"checkAnswer(4, 'C', 'C')\">C. \u5beb\u5728 <code>instructions<\/code> \u9019\u6bb5\u6587\u5b57\u88e1\uff0c\u4f54\u4f4d\u7b26\u6703\u88ab\u6bcf\u984c\u7684 inputs\/outputs \u4ee3\u63db<\/li>\n        <li onclick=\"checkAnswer(4, 'D', 'C')\">D. MLflow \u81ea\u52d5\u7522\u751f\uff0c\u958b\u767c\u8005\u4e0d\u80fd\u4fee\u6539<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb4\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q5\">\n      <h3>5. \u6587\u7ae0\u8aaa\u8981\u7528\u8a55\u4f30\u4f86\u300c\u6bd4\u8f03\u4e0d\u540c prompt \u6216\u6a21\u578b\u7248\u672c\u300d\uff0c\u95dc\u9375\u524d\u63d0\u662f\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(5, 'A', 'D')\">A. \u6bcf\u500b\u7248\u672c\u8981\u7528\u4e0d\u540c\u7684\u8a55\u4f30\u96c6\uff0c\u624d\u516c\u5e73<\/li>\n        <li onclick=\"checkAnswer(5, 'B', 'D')\">B. \u53ea\u8981\u6709\u4e00\u500b\u7248\u672c\u8dd1\u904e evaluate \u5c31\u80fd\u6bd4\u8f03<\/li>\n        <li onclick=\"checkAnswer(5, 'C', 'D')\">C. \u8981\u628a\u6a19\u6e96\u7b54\u6848\u653e\u9032 inputs \u8b93\u5206\u6578\u66f4\u9ad8<\/li>\n        <li onclick=\"checkAnswer(5, 'D', 'D')\">D. \u5169\u6b21\u8a55\u4f30\u8981\u7528\u540c\u4e00\u4efd data \u548c\u540c\u4e00\u7d44 scorers\uff0c\u53ea\u63db predict_fn\uff0c\u5206\u6578\u624d\u53ef\u6bd4<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb5\"><\/div>\n    <\/div>\n\n    <div class=\"score\" id=\"score\" style=\"display:none;\"><\/div>\n<script>\n\n      let answered = 0;\n      let correct = 0;\n      const total = 5;\n      const explanations = {\n        1: \"LLM \u7684\u8f38\u51fa\u6c92\u6709\u552f\u4e00\u6b63\u89e3\uff0c\u540c\u4e00\u984c\u7684\u597d\u7b54\u6848\u6709\u7121\u9650\u591a\u7a2e\uff1bexact match \u6703\u628a\u5927\u91cf\u597d\u7b54\u6848\u5224\u932f\uff0cBLEU\/ROUGE \u53ea\u6e2c\u300c\u7528\u5b57\u50cf\u4e0d\u50cf\u300d\uff0c\u6e2c\u4e0d\u51fa\u901a\u4e0d\u901a\u3001\u6709\u6c92\u6709\u552c\u721b\uff0c\u6240\u4ee5\u624d\u9700\u8981 LLM-as-a-Judge\u3002\",\n        2: \"inputs \u662f\u6703\u88ab\u9935\u9032 predict_fn \u7684\u8f38\u5165\uff08\u984c\u76ee\u3001context\uff09\uff1bexpectations \u662f\u4f60\u5c0d\u9019\u984c\u7684\u671f\u671b\uff08\u5982 expected_response \u6a19\u6e96\u7b54\u6848\uff09\u3002\u6a19\u6e96\u7b54\u6848\u53ea\u8a72\u653e\u5728 expectations\u3002\",\n        3: \"Correctness \u8981\u62ff\u56de\u7b54\u6bd4\u5c0d expected_response\uff0c\u6240\u4ee5\u9700\u8981\u6a19\u6e96\u7b54\u6848\uff1bRelevanceToQuery\u3001RetrievalGroundedness \u53ea\u770b\u984c\u76ee\/\u56de\u7b54\/context \u4e4b\u9593\u5408\u4e0d\u5408\u7406\uff0c\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\u3002\u9019\u6c7a\u5b9a\u4e86\u4f60\u8981\u4e0d\u8981\u82b1\u4eba\u529b\u6e96\u5099\u6a19\u8a3b\u3002\",\n        4: \"\u5224\u6e96\uff08rubric\uff09\u5c31\u662f instructions \u9019\u6bb5\u6587\u5b57\uff0c\u4f60\u5728\u88e1\u9762\u5beb\u6e05\u695a\u4ec0\u9ebc\u7b97 pass\u3001\u4ec0\u9ebc\u7b97 fail\uff1b{{ inputs }} \/ {{ outputs }} \u4f54\u4f4d\u7b26\u6703\u5728\u6bcf\u984c\u88ab\u81ea\u52d5\u4ee3\u63db\u3002judge \u597d\u4e0d\u597d\u7528\u5168\u770b\u9019\u6bb5\u5beb\u5f97\u5920\u4e0d\u5920\u5177\u9ad4\u3002\",\n        5: \"\u6bd4\u8f03\u7684\u524d\u63d0\u662f\u8b8a\u56e0\u55ae\u4e00\uff1a\u56fa\u5b9a\u540c\u4e00\u4efd data \u548c\u540c\u4e00\u7d44 scorers\uff0c\u53ea\u63db predict_fn\u3002\u9019\u6a23\u5169\u500b run \u7684\u5206\u6578\u624d\u53ef\u6bd4\uff0c\u4f60\u624d\u80fd\u8aaa\u300cv2 \u7684 correctness \u5f9e 70% \u5347\u5230 85%\u300d\u3002\",\n      };\n\n      function checkAnswer(qNum, selected, correctAns) {\n        const question = document.getElementById(\"q\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n        question.classList.add(\"answered\");\n        answered++;\n        const options = question.querySelectorAll(\".options li\");\n        options.forEach((opt) => {\n          opt.classList.add(\"disabled\");\n          if (opt.textContent.startsWith(correctAns + \".\")) {\n            opt.classList.add(\"show-answer\");\n          }\n        });\n        const feedback = document.getElementById(\"fb\" + qNum);\n        if (selected === correctAns) {\n          correct++;\n          feedback.className = \"feedback correct\";\n          feedback.innerHTML = \"\u2713 \u6b63\u78ba\uff01\" + explanations[qNum];\n        } else {\n          feedback.className = \"feedback wrong\";\n          feedback.innerHTML = \"\u2717 \u932f\u8aa4\u3002\u6b63\u78ba\u7b54\u6848\u662f \" + correctAns + \"\u3002\" + explanations[qNum];\n          options.forEach((opt) => {\n            if (opt.textContent.startsWith(selected + \".\")) {\n              opt.classList.add(\"wrong\");\n            }\n          });\n        }\n        if (answered === total) {\n          const scoreDiv = document.getElementById(\"score\");\n          scoreDiv.style.display = \"block\";\n          scoreDiv.innerHTML =\n            \"\u6e2c\u9a57\u5b8c\u6210\uff01\u5f97\u5206\uff1a\" + correct + \" \/ \" + total +\n            (correct === total ? \" \ud83c\udf89 \u5168\u5c0d\uff01\" : correct >= Math.ceil(total * 0.6) ? \" \ud83d\udc4d \u4e0d\u932f\uff01\" : \" \ud83d\udcda \u518d\u8907\u7fd2\u4e00\u4e0b\u5427\");\n        }\n      }\n    \n<\/script>\n<\/div>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u7cfb\u5217\u7b2c 2 \u7bc7\uff0c\u5171 4 \u7bc7\uff5c\u96e3\u5ea6\uff1aL3-\u719f\u7df4<br>\u524d\u7f6e\uff1a\u5df2\u8b80 #01\uff08Tracing \u5165\u9580\uff09\uff0c\u4e86\u89e3 trace \u6982\u5ff5\uff1b\u6709\u57fa\u672c Python \u8207 LLM API \u7d93\u9a57<\/p>\n<\/blockquote>\n\n\n\n<h2 class=\"wp-block-heading\">\u4e00\u53e5\u8a71\u8aaa\u660e<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">LLM \u7684\u8f38\u51fa\u6c92\u6709\u6a19\u6e96\u7b54\u6848\uff0c\u6240\u4ee5\u6211\u5011\u8acb\u300c\u53e6\u4e00\u500b LLM \u7576\u8a55\u5be9\u300d\u4f86\u6253\u5206\uff0cMLflow \u628a\u9019\u6574\u5957\u5305\u6210 <code>mlflow.genai.evaluate<\/code>\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u70ba\u4ec0\u9ebc LLM \u8f38\u51fa\u4e0d\u80fd\u7528\u50b3\u7d71 metric<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u50b3\u7d71 ML \u8a55\u4f30\u5f88\u55ae\u7d14\uff1a\u5206\u985e\u554f\u984c\u6bd4\u5c0d label \u7b97 accuracy\uff0c\u56de\u6b78\u554f\u984c\u7b97 RMSE\u3002\u6709\u6a19\u6e96\u7b54\u6848\uff0c\u4e00\u500b <code>==<\/code> \u5c31\u5206\u52dd\u8ca0\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">LLM \u4e0d\u662f\u9019\u6a23\u3002\u4f60\u554f\u300c\u5e6b\u6211\u628a\u9019\u5c01\u4fe1\u5beb\u5f97\u66f4\u5ba2\u6c23\u300d\uff0c\u6b63\u78ba\u7b54\u6848\u6709\u7121\u9650\u591a\u7a2e\u3002\u7528\u5b57\u4e32\u6bd4\u5c0d\uff08exact match\uff09\u6703\u628a 99% \u7684\u597d\u7b54\u6848\u5224\u6210\u932f\uff1b\u7528 BLEU\u3001ROUGE \u9019\u7a2e\u6bd4\u5c0d\u300c\u5b57\u91cd\u758a\u300d\u7684\u6307\u6a19\uff0c\u4e5f\u53ea\u662f\u5728\u6e2c\u300c\u8ddf\u53c3\u8003\u7b54\u6848\u7528\u5b57\u50cf\u4e0d\u50cf\u300d\uff0c\u6e2c\u4e0d\u51fa\u300c\u9019\u53e5\u8a71\u5230\u5e95\u901a\u4e0d\u901a\u3001\u6709\u6c92\u6709\u552c\u721b\u300d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6240\u4ee5\u73fe\u5728\u7684\u4e3b\u6d41\u505a\u6cd5\u662f <strong>LLM-as-a-Judge<\/strong>\uff1a\u62ff\u4e00\u500b LLM \u7576\u8a55\u5be9\uff0c\u7d66\u5b83\u300c\u984c\u76ee\u3001AI \u7684\u56de\u7b54\u3001\uff08\u53ef\u9078\u7684\uff09\u6a19\u6e96\u7b54\u6848\u300d\uff0c\u8acb\u5b83\u7167\u4e00\u4efd\u5224\u6e96\uff08rubric\uff09\u6253\u5206\u6216\u7d66 pass\/fail\u3002MLflow \u7684 GenAI \u8a55\u4f30\u5c31\u662f\u628a\u9019\u4ef6\u4e8b\u6a19\u6e96\u5316\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f5c\u70ba Vibe Coder\uff0c\u4f60\u7684\u4efb\u52d9\u4e0d\u662f\u5f9e\u96f6\u5be6\u4f5c\u8a55\u5be9\u908f\u8f2f\u2014\u2014\u90a3\u662f <code>mlflow.genai<\/code> \u5e6b\u4f60\u505a\u7684\u3002\u4f60\u7684\u4efb\u52d9\u662f<strong>\u770b\u61c2 AI \u5e6b\u4f60\u5beb\u7684\u8a55\u4f30\u8173\u672c\u5728\u8a55\u4ec0\u9ebc\u3001\u5224\u6e96\u56b4\u4e0d\u56b4\u3001\u7d50\u679c\u80fd\u4e0d\u80fd\u4fe1<\/strong>\u3002\u9019\u7bc7\u5c31\u662f\u6559\u4f60\u8b80\u9019\u7a2e code\u3002<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-1\" data-shcb-language-name=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\">\u2705 \u5fc5\u770b\u61c2\uff08\u9019\u7bc7\u7684\u6838\u5fc3\uff09\uff1a\n- \u8a55\u4f30\u8cc7\u6599\u96c6\u9577\u4ec0\u9ebc\u6a23\uff08inputs \/ expectations\uff09\n- mlflow.genai.evaluate(...) \u4e09\u500b\u95dc\u9375\u53c3\u6578\n- \u5167\u5efa scorer\uff08Correctness \/ RelevanceToQuery \/ Guidelines\u2026\uff09\n- \u81ea\u8a02 judge\uff08make_judge\uff09\u7684\u5224\u6e96\u5beb\u5728\u54ea\n\n\ud83d\udccc \u77e5\u9053\u5c31\u597d\uff08\u9047\u5230\u518d\u67e5\uff09\uff1a\n- \u628a\u65e2\u6709 trace \u6488\u51fa\u4f86\u7576\u8a55\u4f30\u8cc7\u6599\n- \u5728 MLflow UI \u6bd4\u8f03\u5169\u500b run\n\n\ud83d\udd70\ufe0f \u8a8d\u5f97\u5c31\u597d\uff08\u820a\u5beb\u6cd5\uff0c\u7dad\u8b77\u820a\u5c08\u6848\u6703\u770b\u5230\uff09\uff1a\n- mlflow.evaluate(model_type=<span class=\"hljs-string\">\"question-answering\"<\/span>)\uff08\u820a\u7248\u8a55\u4f30\u5165\u53e3\uff09\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-1\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">JavaScript<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">javascript<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">30 \u79d2\u7bc4\u4f8b\uff1a\u4e00\u500b\u6700\u5c0f\u7684\u8a55\u4f30<\/h2>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\"><span><code class=\"hljs\">uv add mlflow openai\n<\/code><\/span><\/pre><\/div>\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-2\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">import mlflow\nfrom mlflow.genai.scorers import Correctness\n\n<span class=\"hljs-comment\"># 1. \u8a55\u4f30\u8cc7\u6599\uff1a\u4e00\u7b46\u4e00\u7b46\u7684\u300c\u984c\u76ee + \u671f\u671b\u300d<\/span>\ndata = &#91;\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"MLflow \u7684 tracking server \u9810\u8a2d port \u662f\u5e7e\u865f\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {<span class=\"hljs-string\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"5000\"<\/span>},\n    },\n]\n\n<span class=\"hljs-comment\"># 2. \u88ab\u8a55\u4f30\u7684\u6771\u897f\uff1a\u4e00\u500b\u5403 inputs\u3001\u5410\u56de\u7b54\u7684\u51fd\u5f0f<\/span>\ndef my_app(question: str) -&gt; str:\n    <span class=\"hljs-keyword\">return<\/span> call_my_llm(question)   <span class=\"hljs-comment\"># \u4f60\u7684 LLM \u61c9\u7528<\/span>\n\n<span class=\"hljs-comment\"># 3. \u8dd1\u8a55\u4f30<\/span>\nresults = mlflow.genai.evaluate(\n    data=data,\n    predict_fn=my_app,\n    scorers=&#91;Correctness()],       <span class=\"hljs-comment\"># \u7528\u5167\u5efa\u7684\u300c\u6b63\u78ba\u6027\u300d\u8a55\u5be9<\/span>\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-2\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u6bb5\u4ee3\u78bc\u505a\u4e86\u4ec0\u9ebc<\/strong>\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n\n<li><code>data<\/code>\uff1a\u6e96\u5099\u4e00\u4efd\u8a55\u4f30\u96c6\uff0c\u6bcf\u7b46\u6709 <code>inputs<\/code>\uff08\u9935\u7d66\u4f60 app \u7684\u984c\u76ee\uff09\u548c <code>expectations<\/code>\uff08\u4f60\u671f\u671b\u5b83\u7b54\u6210\u600e\u6a23\uff09<\/li>\n\n\n<li><code>predict_fn<\/code>\uff1a\u6307\u5411\u4f60\u8981\u88ab\u6253\u5206\u7684\u61c9\u7528\uff0cMLflow \u6703\u628a\u6bcf\u7b46 <code>inputs<\/code> \u9935\u9032\u53bb\u62ff\u56de\u7b54<\/li>\n\n\n<li><code>scorers<\/code>\uff1a\u6307\u5b9a\u8a55\u5be9\u3002<code>Correctness()<\/code> \u662f\u5167\u5efa\u7684 LLM judge\uff0c\u6703\u62ff\u300c\u56de\u7b54\u300d\u5c0d\u300cexpected_response\u300d\u5224\u5c0d\u932f<\/li>\n\n\n<li>\u8dd1\u5b8c\uff0c\u7d50\u679c\u9032 MLflow\uff0c\u53ef\u4ee5\u5728 UI \u770b\u6bcf\u4e00\u984c\u7684\u5206\u6578\u548c\u8a55\u5be9\u7406\u7531<\/li>\n\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u6838\u5fc3\u6982\u5ff5\u7ffb\u8b6f<\/h2>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u4f60\u6703\u770b\u5230<\/th>\n<th>\u610f\u601d<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>inputs<\/code><\/td>\n<td>\u9935\u7d66\u4f60 app \u7684\u8f38\u5165\uff08\u984c\u76ee\u3001context\u2026\uff09\uff0c\u4e00\u500b dict<\/td>\n<\/tr>\n<tr>\n<td><code>expectations<\/code><\/td>\n<td>\u4f60\u5c0d\u9019\u984c\u7684\u671f\u671b\uff08\u6a19\u6e96\u7b54\u6848\u3001\u8a72\u9075\u5b88\u7684\u898f\u5247\uff09\uff0c\u8a55\u5be9\u6703\u53c3\u8003<\/td>\n<\/tr>\n<tr>\n<td><code>predict_fn<\/code><\/td>\n<td>\u88ab\u8a55\u4f30\u7684\u61c9\u7528\u51fd\u5f0f\uff0c\u5403 inputs \u5410 outputs<\/td>\n<\/tr>\n<tr>\n<td><code>scorers=[...]<\/code><\/td>\n<td>\u4e00\u4e32\u8a55\u5be9\u3002\u53ef\u4ee5\u6df7\u7528\u5167\u5efa scorer \u548c\u81ea\u8a02 judge<\/td>\n<\/tr>\n<tr>\n<td><code>Correctness()<\/code><\/td>\n<td>\u5167\u5efa LLM judge\uff1a\u56de\u7b54\u5c0d\u4e0d\u5c0d\uff08\u6bd4\u5c0d expected_response\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>RelevanceToQuery()<\/code><\/td>\n<td>\u5167\u5efa\uff1a\u56de\u7b54\u6709\u6c92\u6709\u5207\u984c\uff08\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>RetrievalGroundedness()<\/code><\/td>\n<td>\u5167\u5efa\uff1a\u56de\u7b54\u6709\u6c92\u6709\u300c\u6839\u64da context\u300d\u800c\u4e0d\u662f\u81ea\u5df1\u778e\u7de8\uff08\u6e2c RAG \u5e7b\u89ba\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>Guidelines(...)<\/code><\/td>\n<td>\u5167\u5efa\uff1a\u4f60\u7528\u767d\u8a71\u5beb\u4e00\u689d\u898f\u5247\uff0cjudge \u5224\u6709\u6c92\u6709\u9075\u5b88<\/td>\n<\/tr>\n<tr>\n<td><code>make_judge(...)<\/code><\/td>\n<td>\u81ea\u5df1\u9020\u4e00\u500b judge\uff0c\u5224\u6e96\uff08rubric\uff09\u4f60\u81ea\u5df1\u5beb<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e00\u500b\u95dc\u9375\u76f4\u89ba\uff1a<strong>\u6709\u4e9b scorer \u9700\u8981 <code>expectations<\/code>\uff08\u6a19\u6e96\u7b54\u6848\uff09\uff0c\u6709\u4e9b\u4e0d\u9700\u8981<\/strong>\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li><code>Correctness<\/code> \u8981\u6a19\u6e96\u7b54\u6848\uff08\u62ff\u56de\u7b54\u6bd4\u5c0d expected_response\uff09<\/li>\n\n\n<li><code>RelevanceToQuery<\/code>\u3001<code>RetrievalGroundedness<\/code> \u4e0d\u7528\u6a19\u6e96\u7b54\u6848\uff08\u53ea\u770b\u984c\u76ee\u3001\u56de\u7b54\u3001context \u4e4b\u9593\u5408\u4e0d\u5408\u7406\uff09<\/li>\n\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u5f88\u91cd\u8981\uff0c\u56e0\u70ba\u6e96\u5099\u6a19\u6e96\u7b54\u6848\u5f88\u8cb4\u3002\u770b\u5230\u4e00\u4efd\u8a55\u4f30\u8173\u672c\u53ea\u7528\u4e86\u300c\u4e0d\u9700\u6a19\u6e96\u7b54\u6848\u300d\u7684 scorer\uff0c\u4f60\u5c31\u77e5\u9053\u5b83\u7701\u4e86\u4eba\u5de5\u6a19\u8a3b\uff0c\u4f46\u4e5f\u5c31\u8a55\u4e0d\u4e86\u300c\u7b54\u6848\u6b63\u4e0d\u6b63\u78ba\u300d\u9019\u4ef6\u4e8b\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u8a55\u4f30\u8cc7\u6599\u96c6\uff1a\u4e00\u7b46\u8cc7\u6599\u9577\u600e\u6a23<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u662f\u6574\u500b\u8a55\u4f30\u7684\u5730\u57fa\uff0c\u4e00\u5b9a\u8981\u770b\u61c2\u3002\u6700\u5b8c\u6574\u7684\u4e00\u7b46\u9577\u9019\u6a23\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-3\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">{\n    <span class=\"hljs-string\">\"inputs\"<\/span>: {                              <span class=\"hljs-comment\"># \u9935\u7d66 app \u7684\u8f38\u5165<\/span>\n        <span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u9000\u8ca8\u8981\u5e7e\u5929\u5167\uff1f\"<\/span>,\n        <span class=\"hljs-string\">\"context\"<\/span>: <span class=\"hljs-string\">\"\u672c\u5e97\u9000\u8ca8\u653f\u7b56\u70ba 7 \u5929\u5167\u2026\"<\/span>,   <span class=\"hljs-comment\"># RAG \u5834\u666f\u6703\u6709\u6aa2\u7d22\u5230\u7684\u8cc7\u6599<\/span>\n    },\n    <span class=\"hljs-string\">\"expectations\"<\/span>: {                        <span class=\"hljs-comment\"># \u4f60\u7684\u671f\u671b\uff08\u8a55\u5be9\u53c3\u8003\u7528\uff09<\/span>\n        <span class=\"hljs-string\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"7 \u5929\u5167\u53ef\u9000\u8ca8\"<\/span>,  <span class=\"hljs-comment\"># \u6a19\u6e96\u7b54\u6848<\/span>\n        <span class=\"hljs-string\">\"expected_facts\"<\/span>: &#91;<span class=\"hljs-string\">\"7 \u5929\"<\/span>],          <span class=\"hljs-comment\"># \u5fc5\u9808\u63d0\u5230\u7684\u4e8b\u5be6\u9ede<\/span>\n    },\n}\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-3\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9010\u6b04\u7ffb\u8b6f<\/strong>\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-4\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-string\">\"inputs\"<\/span>                <span class=\"hljs-comment\"># \u9019\u984c\u8981\u554f app \u4ec0\u9ebc\uff08\u6703\u88ab\u9935\u9032 predict_fn\uff09<\/span>\n  <span class=\"hljs-string\">\"question\"<\/span>            <span class=\"hljs-comment\"># \u4f7f\u7528\u8005\u7684\u554f\u984c<\/span>\n  <span class=\"hljs-string\">\"context\"<\/span>            <span class=\"hljs-comment\"># \u6aa2\u7d22\u5230\u7684\u53c3\u8003\u8cc7\u6599\uff08RAG \u624d\u6709\uff09<\/span>\n<span class=\"hljs-string\">\"expectations\"<\/span>          <span class=\"hljs-comment\"># \u9019\u984c\u300c\u7b54\u5c0d\u300d\u9577\u4ec0\u9ebc\u6a23<\/span>\n  <span class=\"hljs-string\">\"expected_response\"<\/span>   <span class=\"hljs-comment\"># \u5b8c\u6574\u7684\u7406\u60f3\u7b54\u6848<\/span>\n  <span class=\"hljs-string\">\"expected_facts\"<\/span>      <span class=\"hljs-comment\"># \u53ea\u8981\u7b54\u6848\u5305\u542b\u9019\u4e9b\u95dc\u9375\u4e8b\u5be6\u5c31\u7b97\u5c0d\uff08\u6bd4\u9010\u5b57\u6bd4\u5c0d\u5bec\u9b06\uff09<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-4\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\">\u4f60\u4e0d\u7528\u6bcf\u500b\u6b04\u4f4d\u90fd\u586b\u3002\u6700\u7701\u7684\u7248\u672c\u53ea\u6709 <code>inputs<\/code>\uff08\u914d\u4e0d\u9700\u6a19\u6e96\u7b54\u6848\u7684 scorer\uff09\uff1b\u8981\u8a55\u6b63\u78ba\u6027\u624d\u9700\u8981 <code>expectations<\/code>\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u628a trace \u8f49\u6210\u8a55\u4f30\u8cc7\u6599\uff08\u627f\u63a5 #01\uff09<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">#01 \u6559\u4f60\u7684 trace\uff0c\u5176\u5be6\u5c31\u662f\u73fe\u6210\u7684\u8a55\u4f30\u7d20\u6750\u2014\u2014\u88e1\u9762\u5df2\u7d93\u6709\u771f\u5be6\u4f7f\u7528\u8005\u554f\u4e86\u4ec0\u9ebc\u3001\u4f60\u7684 app \u56de\u4e86\u4ec0\u9ebc\u3002MLflow \u53ef\u4ee5\u628a trace \u6488\u51fa\u4f86\u76f4\u63a5\u7576 <code>data<\/code>\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-5\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u628a\u904e\u53bb\u8a18\u9304\u4e0b\u4f86\u7684 trace \u6488\u51fa\u4f86\u7576\u8a55\u4f30\u96c6<\/span>\ntraces = mlflow.search_traces(\n    experiment_ids=&#91;<span class=\"hljs-string\">\"123\"<\/span>],\n    max_results=<span class=\"hljs-number\">50<\/span>,\n)\n\nresults = mlflow.genai.evaluate(\n    data=traces,          <span class=\"hljs-comment\"># \u76f4\u63a5\u9935 trace\uff0c\u4e0d\u7528\u624b\u5beb inputs<\/span>\n    scorers=&#91;RelevanceToQuery()],\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-5\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u5728\u5e79\u561b<\/strong>\uff1a\u4e0d\u7528\u81ea\u5df1\u624b\u523b\u8a55\u4f30\u96c6\uff0c\u76f4\u63a5\u62ff\u7dda\u4e0a\u771f\u5be6\u6d41\u91cf\u4f86\u8a55\u3002MLflow \u8a8d\u5f97 trace \u683c\u5f0f\uff0c\u6703\u81ea\u52d5\u62bd\u51fa inputs \/ outputs\u3002\u9019\u4e5f\u662f\u70ba\u4ec0\u9ebc #01 \u8981\u4f60\u5148\u5b78\u6703 tracing\u2014\u2014\u6c92\u6709 trace \u5c31\u6c92\u6709\u771f\u5be6\u8a55\u4f30\u8cc7\u6599\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5167\u5efa scorers\uff1aAI \u6700\u5e38\u7528\u9019\u5e7e\u500b<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u770b\u8a55\u4f30\u8173\u672c\u6642\uff0c\u5148\u770b\u5b83 import \u4e86\u54ea\u4e9b scorer\uff0c\u5c31\u5927\u6982\u77e5\u9053\u5b83\u5728\u8a55\u4ec0\u9ebc\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u7528\u6cd5 1\uff1aGuidelines \u2014\u2014 \u7528\u767d\u8a71\u5beb\u898f\u5247<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u662f\u6700\u597d\u61c2\u3001\u4e5f\u6700\u5e38\u88ab AI \u62ff\u4f86\u7528\u7684\u4e00\u7a2e\u3002\u4f60\u7528\u81ea\u7136\u8a9e\u8a00\u5beb\u4e00\u689d\u898f\u5247\uff0cjudge \u5e6b\u4f60\u5224\u6709\u6c92\u6709\u9075\u5b88\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-6\" data-shcb-language-name=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\"><span class=\"hljs-keyword\">from<\/span> mlflow.genai.scorers <span class=\"hljs-keyword\">import<\/span> Guidelines\n\ntone = Guidelines(\n    name=<span class=\"hljs-string\">\"polite_tone\"<\/span>,\n    guidelines=<span class=\"hljs-string\">\"\u56de\u7b54\u5fc5\u9808\u4fdd\u6301\u79ae\u8c8c\u3001\u5c08\u696d\uff0c\u4e0d\u80fd\u51fa\u73fe\u60c5\u7dd2\u6027\u6216\u8cac\u5099\u4f7f\u7528\u8005\u7684\u5b57\u773c\u3002\"<\/span>,\n)\n\nresults = mlflow.genai.evaluate(\n    data=data,\n    predict_fn=my_app,\n    scorers=&#91;tone],\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-6\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">JavaScript<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">javascript<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7ffb\u8b6f<\/strong>\uff1a\u300c\u8acb\u4e00\u500b LLM \u8b80\u6bcf\u500b\u56de\u7b54\uff0c\u5224\u65b7\u5b83\u6709\u6c92\u6709\u505a\u5230\u300e\u79ae\u8c8c\u5c08\u696d\u300f\u9019\u689d\u898f\u5247\uff0c\u7d66 pass \u6216 fail \u52a0\u7406\u7531\u3002\u300d\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\uff0c\u56e0\u70ba\u5224\u7684\u662f\u300c\u98a8\u683c\u6709\u6c92\u6709\u5b88\u898f\u77e9\u300d\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u7528\u6cd5 2\uff1a\u591a\u500b scorer \u4e00\u8d77\u8a55<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u771f\u5be6\u8173\u672c\u901a\u5e38\u4e00\u6b21\u639b\u597d\u5e7e\u500b\u8a55\u5be9\uff0c\u5f9e\u4e0d\u540c\u89d2\u5ea6\u6253\u5206\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-7\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.scorers import (\n    Correctness, RelevanceToQuery, RetrievalGroundedness,\n)\n\nresults = mlflow.genai.evaluate(\n    data=data,\n    predict_fn=rag_app,\n    scorers=&#91;\n        Correctness(),            <span class=\"hljs-comment\"># \u7b54\u5c0d\u55ce\uff08\u8981 expected_response\uff09<\/span>\n        RelevanceToQuery(),       <span class=\"hljs-comment\"># \u5207\u984c\u55ce<\/span>\n        RetrievalGroundedness(),  <span class=\"hljs-comment\"># \u6709\u6c92\u6709\u6839\u64da\u6aa2\u7d22\u8cc7\u6599\u3001\u9084\u662f\u5728\u5e7b\u89ba<\/span>\n    ],\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-7\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7ffb\u8b6f<\/strong>\uff1a\u9019\u662f\u5178\u578b\u7684 RAG \u8a55\u4f30\u7d44\u5408\u2014\u2014\u540c\u6642\u6e2c\u300c\u6b63\u78ba\u3001\u5207\u984c\u3001\u4e0d\u4e82\u7de8\u300d\u3002\u7d50\u679c\u8868\u683c\u6bcf\u500b scorer \u4e00\u6b04\uff0c\u4f60\u80fd\u770b\u51fa\u300c\u7b54\u5c0d\u4f46\u96e2\u984c\u300d\u6216\u300c\u5207\u984c\u4f46\u5728\u552c\u721b\u300d\u9019\u7a2e\u7d30\u7bc0\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u81ea\u8a02 judge\uff1a\u5224\u6e96\uff08rubric\uff09\u81ea\u5df1\u5beb<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5167\u5efa scorer \u4e0d\u5920\u7528\u6642\uff0cAI \u6703\u5e6b\u4f60\u7528 <code>make_judge<\/code> \u9020\u4e00\u500b\u3002\u9019\u662f\u8a55\u4f30\u8173\u672c\u88e1\u6700\u9700\u8981\u76ef\u7dca\u7684\u90e8\u5206\uff0c\u56e0\u70ba<strong>\u5224\u6e96\u7684\u9b06\u7dca\u5168\u5728\u9019\u6bb5\u6587\u5b57\u88e1<\/strong>\u3002<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-8\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.judges import make_judge\n\nconciseness = make_judge(\n    name=<span class=\"hljs-string\">\"conciseness\"<\/span>,\n    instructions=(\n        <span class=\"hljs-string\">\"\u8a55\u4f30\u9019\u500b\u56de\u7b54\u662f\u5426\u7c21\u6f54\u3002\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u554f\u984c\uff1a{{ inputs }}\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u56de\u7b54\uff1a{{ outputs }}\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u5982\u679c\u56de\u7b54\u76f4\u63a5\u3001\u6c92\u6709\u5ee2\u8a71\u3001\u6c92\u6709\u91cd\u8907\uff0c\u8a55\u70ba 'pass'\uff1b\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u5982\u679c\u5197\u9577\u3001\u7e5e\u5708\u5b50\u3001\u91cd\u8907\u540c\u6a23\u7684\u610f\u601d\uff0c\u8a55\u70ba 'fail'\u3002\"<\/span>\n    ),\n    model=<span class=\"hljs-string\">\"openai:\/gpt-4o\"<\/span>,       <span class=\"hljs-comment\"># \u7528\u54ea\u500b\u6a21\u578b\u7576\u8a55\u5be9<\/span>\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-8\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9010\u6bb5\u7ffb\u8b6f<\/strong>\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-9\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">name=<span class=\"hljs-string\">\"conciseness\"<\/span>        <span class=\"hljs-comment\"># \u9019\u500b\u8a55\u5be9\u7684\u540d\u5b57\uff08\u6703\u8b8a\u6210\u7d50\u679c\u8868\u683c\u7684\u6b04\u540d\uff09<\/span>\ninstructions=<span class=\"hljs-string\">\"...\"<\/span>        <span class=\"hljs-comment\"># \u5224\u6e96\u672c\u9ad4\uff1a\u544a\u8a34\u8a55\u5be9 LLM \u600e\u9ebc\u6253\u5206<\/span>\n  <span class=\"hljs-string\">\"{{ inputs }}\"<\/span>          <span class=\"hljs-comment\"># \u4f54\u4f4d\u7b26\uff0c\u8dd1\u7684\u6642\u5019\u63db\u6210\u8a72\u984c\u7684\u8f38\u5165<\/span>\n  <span class=\"hljs-string\">\"{{ outputs }}\"<\/span>         <span class=\"hljs-comment\"># \u4f54\u4f4d\u7b26\uff0c\u63db\u6210 app \u7684\u56de\u7b54<\/span>\n  <span class=\"hljs-string\">\"\u8a55\u70ba 'pass' \/ 'fail'\"<\/span>  <span class=\"hljs-comment\"># \u660e\u78ba\u7684\u8f38\u51fa\u683c\u5f0f\uff08judge \u8981\u56de\u9019\u500b\uff09<\/span>\nmodel=<span class=\"hljs-string\">\"openai:\/gpt-4o\"<\/span>    <span class=\"hljs-comment\"># \u6307\u5b9a\u7576\u8a55\u5be9\u7684\u6a21\u578b\uff08\u8d8a\u5f37\u7684\u6a21\u578b\u5224\u5f97\u8d8a\u6e96\uff0c\u4f46\u8d8a\u8cb4\uff09<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-9\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u5728\u5e79\u561b<\/strong>\uff1a\u4f60\u5728 <code>instructions<\/code> \u88e1\u5beb\u6e05\u695a\u300c\u4ec0\u9ebc\u7b97\u597d\u3001\u4ec0\u9ebc\u7b97\u58de\u300d\uff0c<code>{{ inputs }}<\/code> \/ <code>{{ outputs }}<\/code> \u9019\u4e9b\u4f54\u4f4d\u7b26\u6703\u5728\u6bcf\u4e00\u984c\u88ab\u81ea\u52d5\u4ee3\u63db\u3002\u9019\u5c31\u662f\u4f60\u7684 rubric\u2014\u2014judge \u597d\u4e0d\u597d\u7528\uff0c\u5168\u770b\u9019\u6bb5\u5beb\u5f97\u5920\u4e0d\u5920\u5177\u9ad4\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5224\u6e96\u8a2d\u8a08\u7684\u91cd\u9ede\uff1a<strong>\u8d8a\u5177\u9ad4\u8d8a\u53ef\u9760<\/strong>\u3002\u300c\u8a55\u4f30\u56de\u7b54\u54c1\u8cea\u300d\u9019\u7a2e\u6a21\u7cca\u6307\u4ee4\u6703\u8b93 judge \u4e82\u7d66\u5206\uff1b\u300c\u6709\u6c92\u6709\u63d0\u5230\u9000\u8ca8\u5929\u6578\u3001\u8a9e\u6c23\u79ae\u4e0d\u79ae\u8c8c\u300d\u9019\u7a2e\u53ef\u6aa2\u67e5\u7684\u689d\u4ef6\u624d\u7a69\u5b9a\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u8b80\u61c2\u8a55\u4f30\u7d50\u679c<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u8dd1\u5b8c <code>evaluate<\/code> \u6703\u56de\u4e00\u500b\u7d50\u679c\u7269\u4ef6\uff0c\u4e5f\u6703\u5beb\u9032 MLflow\u3002\u4f60\u6703\u770b\u5230\u985e\u4f3c\u9019\u6a23\u7684\u8868\u683c\uff08\u6bcf\u5217\u4e00\u984c\uff0c\u6bcf\u500b scorer \u4e00\u6b04\uff09\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\"><span><code class=\"hljs\">question              | correctness | relevance | conciseness\n----------------------|-------------|-----------|------------\n\u9000\u8ca8\u8981\u5e7e\u5929\uff1f           | pass        | pass      | fail\n\u904b\u8cbb\u600e\u9ebc\u7b97\uff1f           | fail        | pass      | pass\n<\/code><\/span><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u600e\u9ebc\u8b80<\/strong>\uff1a\u6a6b\u8457\u770b\u55ae\u984c\uff08\u9019\u984c\u54ea\u500b\u9762\u5411\u639b\u4e86\uff09\uff0c\u76f4\u8457\u770b\u8da8\u52e2\uff08\u54ea\u500b scorer \u6574\u9ad4\u504f\u4f4e\uff09\u3002\u4e0a\u9762\u7b2c\u4e00\u5217\u300c\u7b54\u5c0d\u3001\u5207\u984c\uff0c\u4f46\u592a\u56c9\u55e6\u300d\uff1b\u7b2c\u4e8c\u5217\u300c\u5207\u984c\u3001\u7c21\u6f54\uff0c\u4f46\u7b54\u932f\u300d\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u5728 UI \u6bd4\u8f03\u4e0d\u540c prompt \/ \u6a21\u578b\u7248\u672c<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u8a55\u4f30\u6700\u5927\u7684\u50f9\u503c\u662f<strong>\u6bd4\u8f03<\/strong>\u3002\u4f60\u6539\u4e86 prompt \u6216\u63db\u4e86\u6a21\u578b\uff0c\u5404\u8dd1\u4e00\u6b21 evaluate\uff0c\u5c31\u662f\u5169\u500b MLflow run\uff0c\u5728 UI \u4e26\u6392\u770b\u54ea\u500b\u5206\u6578\u9ad8\uff1a<\/p>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-10\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u7248\u672c A<\/span>\nwith mlflow.start_run(run_name=<span class=\"hljs-string\">\"prompt-v1\"<\/span>):\n    mlflow.genai.evaluate(data=data, predict_fn=app_v1, scorers=scorers)\n\n<span class=\"hljs-comment\"># \u7248\u672c B<\/span>\nwith mlflow.start_run(run_name=<span class=\"hljs-string\">\"prompt-v2\"<\/span>):\n    mlflow.genai.evaluate(data=data, predict_fn=app_v2, scorers=scorers)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-10\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u9019\u5728\u5e79\u561b<\/strong>\uff1a\u5169\u6b21\u8a55\u4f30\u7528<strong>\u540c\u4e00\u4efd data \u548c\u540c\u4e00\u7d44 scorers<\/strong>\uff0c\u53ea\u63db <code>predict_fn<\/code>\u3002\u9019\u6a23\u5169\u500b run \u7684\u5206\u6578\u624d\u53ef\u6bd4\u2014\u2014\u4f60\u624d\u80fd\u7406\u76f4\u6c23\u58ef\u8aaa\u300cv2 \u7684 correctness \u5f9e 70% \u5347\u5230 85%\u300d\u3002\u5230 MLflow UI \u52fe\u9078\u5169\u500b run \u5c31\u80fd\u4e26\u6392\u6bd4\u8f03\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udea9 \u7d05\u65d7\uff1a\u8a55\u4f30\u8173\u672c\u6700\u5bb9\u6613\u51fa\u7684\u554f\u984c<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u8a55\u4f30\u9019\u7a2e\u6771\u897f\u6700\u9670\u96aa\u7684\u5730\u65b9\u662f\u2014\u2014<strong>\u5b83\u6c38\u9060\u6703\u7d66\u4f60\u4e00\u500b\u6578\u5b57\uff0c\u4f46\u6578\u5b57\u4e0d\u4e00\u5b9a\u6709\u610f\u7fa9<\/strong>\u3002\u4ee5\u4e0b\u5e7e\u500b\u662f AI \u5e6b\u4f60\u5beb\u8a55\u4f30\u6642\u6700\u5e38\u898b\u7684\u5751\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udea9 \u7d05\u65d7 1\uff1ajudge prompt \u904e\u65bc\u5bec\u9b06<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-11\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">quality = make_judge(\n    name=<span class=\"hljs-string\">\"quality\"<\/span>,\n    instructions=<span class=\"hljs-string\">\"\u5224\u65b7\u9019\u500b\u56de\u7b54\u597d\u4e0d\u597d\u3002{{ outputs }}\"<\/span>,   <span class=\"hljs-comment\"># \ud83d\udea9 \u5224\u6e96\u592a\u6a21\u7cca<\/span>\n    model=<span class=\"hljs-string\">\"openai:\/gpt-4o\"<\/span>,\n)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-11\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc\u5371\u96aa<\/strong>\uff1a\u300c\u597d\u4e0d\u597d\u300d\u6c92\u6709\u53ef\u6aa2\u67e5\u7684\u6a19\u6e96\uff0cjudge \u53ea\u80fd\u6191\u611f\u89ba\uff0c\u7d50\u679c\u5e7e\u4e4e\u5168 pass\uff0c\u4f60\u6703\u8aa4\u4ee5\u70ba\u6a21\u578b\u5f88\u68d2\u3002\u9019\u7a2e\u8a55\u4f30\u7b49\u65bc\u6c92\u8a55\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u628a\u9019\u500b judge \u7684 instructions \u6539\u5177\u9ad4\uff0c\u5217\u51fa\u660e\u78ba\u53ef\u6aa2\u67e5\u7684\u689d\u4ef6\uff08\u4f8b\u5982\u8981\u5305\u542b\u54ea\u4e9b\u8cc7\u8a0a\u3001\u8a9e\u6c23\u8981\u6c42\u3001\u9577\u5ea6\u9650\u5236\uff09\uff0c\u4e26\u8981\u6c42\u5b83\u8f38\u51fa pass\/fail \u52a0\u7406\u7531\u3002\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udea9 \u7d05\u65d7 2\uff1a\u8cc7\u6599\u6d29\u6f0f\uff08\u6a19\u6e96\u7b54\u6848\u6df7\u9032\u8f38\u5165\uff09<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-12\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">data = &#91;{\n    <span class=\"hljs-string\">\"inputs\"<\/span>: {\n        <span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u9000\u8ca8\u5e7e\u5929\uff1f\"<\/span>,\n        <span class=\"hljs-string\">\"answer\"<\/span>: <span class=\"hljs-string\">\"7 \u5929\"<\/span>,          <span class=\"hljs-comment\"># \ud83d\udea9 \u6a19\u6e96\u7b54\u6848\u88ab\u653e\u9032 inputs \u9935\u7d66 app \u4e86<\/span>\n    },\n    <span class=\"hljs-string\">\"expectations\"<\/span>: {<span class=\"hljs-string\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"7 \u5929\"<\/span>},\n}]\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-12\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc\u5371\u96aa<\/strong>\uff1a<code>inputs<\/code> \u662f\u6703\u9935\u9032\u4f60 app \u7684\u6771\u897f\u3002\u628a\u7b54\u6848\u593e\u5728 inputs \u88e1\uff0c\u7b49\u65bc\u8003\u8a66\u628a\u7b54\u6848\u5370\u5728\u8003\u5377\u4e0a\u2014\u2014app \u76f4\u63a5\u6284\uff0c\u8a55\u4f30\u5206\u6578\u865b\u9ad8\uff0c\u4e0a\u7dda\u5c31\u73fe\u5f62\u3002\u6a19\u6e96\u7b54\u6848\u53ea\u8a72\u653e\u5728 <code>expectations<\/code>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u6aa2\u67e5\u6211\u7684\u8a55\u4f30\u96c6\uff0c\u6a19\u6e96\u7b54\u6848\u61c9\u8a72\u53ea\u51fa\u73fe\u5728 expectations\uff0c\u4e0d\u80fd\u51fa\u73fe\u5728 inputs \u88e1\u3002\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udea9 \u7d05\u65d7 3\uff1a\u7528\u88ab\u8a55\u4f30\u7684\u540c\u4e00\u500b\u6a21\u578b\u7576\u8a55\u5be9<\/h3>\n\n\n<div class=\"cbc-code-wrapper\"><div class=\"cbc-code-bar\"><button class=\"cbc-copy-button\">Copy<\/button><\/div><pre class=\"wp-block-code\" aria-describedby=\"shcb-language-13\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">judge = make_judge(name=<span class=\"hljs-string\">\"x\"<\/span>, instructions=<span class=\"hljs-string\">\"...\"<\/span>, model=<span class=\"hljs-string\">\"openai:\/gpt-4o\"<\/span>)\n<span class=\"hljs-comment\"># \u800c my_app \u5167\u90e8\u4e5f\u662f\u7528 gpt-4o \u751f\u6210\u56de\u7b54   \ud83d\udea9 \u7403\u54e1\u517c\u88c1\u5224<\/span>\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-13\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">PHP<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">php<\/span><span class=\"shcb-language__paren\">)<\/span><\/small><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc\u5371\u96aa<\/strong>\uff1a\u540c\u4e00\u500b\u6a21\u578b\u6709\u5171\u540c\u76f2\u9ede\uff0c\u5b83\u770b\u4e0d\u51fa\u81ea\u5df1\u7684\u932f\uff0c\u9084\u53ef\u80fd\u504f\u597d\u300c\u8ddf\u81ea\u5df1\u98a8\u683c\u50cf\u300d\u7684\u7b54\u6848\uff08self-preference bias\uff09\uff0c\u5206\u6578\u6703\u7cfb\u7d71\u6027\u504f\u9ad8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u8a55\u5be9\u6a21\u578b\u548c\u88ab\u8a55\u4f30\u7684\u6a21\u578b\u4e0d\u8981\u7528\u540c\u4e00\u500b\uff0c\u63db\u4e00\u500b\u7368\u7acb\u7684\uff08\u6700\u597d\u662f\u66f4\u5f37\u7684\uff09\u6a21\u578b\u7576 judge\u3002\u300d<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\ud83d\udea9 \u7d05\u65d7 4\uff1a\u53ea\u4fe1 judge\uff0c\u5b8c\u5168\u6c92\u6709\u4eba\u5de5\u62bd\u67e5<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u70ba\u4ec0\u9ebc\u5371\u96aa<\/strong>\uff1aLLM judge \u672c\u8eab\u4e5f\u6703\u51fa\u932f\u3001\u4e5f\u6703\u6709\u504f\u597d\u3002\u5982\u679c\u4f60\u5f9e\u6c92\u4eba\u5de5\u770b\u904e judge \u7684\u5224\u65b7\u5c0d\u4e0d\u5c0d\uff0c\u4f60\u7b49\u65bc\u7528\u4e00\u500b\u6c92\u88ab\u9a57\u8b49\u904e\u7684\u5c3a\u5728\u91cf\u6771\u897f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u8ddf AI \u9019\u6a23\u8aaa<\/strong>\uff1a\u300c\u5e6b\u6211\u62bd 10 \u7b46 judge \u5224 pass \u548c fail \u7684\u7d50\u679c\u5217\u51fa\u4f86\uff0c\u6211\u8981\u4eba\u5de5\u6838\u5c0d judge \u5224\u5f97\u5c0d\u4e0d\u5c0d\u3002\u300d\uff08\u6b63\u5f0f\u505a\u6cd5\u662f\u5148\u8b93 judge \u548c\u4eba\u985e\u6a19\u8a3b\u5c0d\u9f4a\uff0c\u518d\u653e\u5fc3\u5927\u898f\u6a21\u7528\u3002\uff09<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">Vibe Coder \u9a57\u6536\u6aa2\u67e5\u9ede<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u62ff\u5230\u4e00\u4efd AI \u5beb\u7684\u8a55\u4f30\u8173\u672c\uff0c\u7167\u9019\u500b\u6e05\u55ae\u904e\u4e00\u904d\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li>[ ] \u6253\u958b\u8a55\u4f30\u96c6\uff0c\u78ba\u8a8d <strong>\u6a19\u6e96\u7b54\u6848\u53ea\u5728 <code>expectations<\/code><\/strong>\u3001\u6c92\u6df7\u9032 <code>inputs<\/code>\uff08\u9632\u8cc7\u6599\u6d29\u6f0f\uff09<\/li>\n\n\n<li>[ ] \u627e\u5230\u6bcf\u500b <code>make_judge<\/code> \u7684 <code>instructions<\/code>\uff0c\u8b80\u4e00\u904d\u2014\u2014\u5224\u6e96\u662f<strong>\u5177\u9ad4\u53ef\u6aa2\u67e5<\/strong>\u7684\uff0c\u9084\u662f\u300c\u597d\u4e0d\u597d\u300d\u9019\u7a2e\u6a21\u7cca\u8a5e\uff1f<\/li>\n\n\n<li>[ ] \u78ba\u8a8d <strong>judge \u7684 model \u548c\u88ab\u8a55\u4f30\u7684 model \u4e0d\u662f\u540c\u4e00\u500b<\/strong><\/li>\n\n\n<li>[ ] \u8dd1\u4e00\u6b21\u8a55\u4f30\uff0c\u5230 MLflow UI \u6253\u958b\u7d50\u679c\u8868\u683c\uff0c<strong>\u624b\u52d5\u770b 3-5 \u984c<\/strong> judge \u7684\u8a55\u5206\u7406\u7531\uff0c\u5224\u65b7\u5b83\u5224\u5f97\u5408\u4e0d\u5408\u7406<\/li>\n\n\n<li>[ ] \u8981\u6bd4\u8f03\u7248\u672c\u6642\uff0c\u78ba\u8a8d\u5169\u500b run \u7528\u4e86<strong>\u540c\u4e00\u4efd data \u548c\u540c\u4e00\u7d44 scorers<\/strong>\uff08\u5426\u5247\u5206\u6578\u4e0d\u53ef\u6bd4\uff09<\/li>\n\n\n<li>[ ] \u554f AI\uff1a\u300c\u9019\u4efd\u8a55\u4f30\u6709\u6c92\u6709\u9700\u8981\u6a19\u6e96\u7b54\u6848\u7684 scorer\uff1f\u6211\u7684\u8a55\u4f30\u96c6\u6709\u6e96\u5099 expectations \u55ce\uff1f\u300d<\/li>\n\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u770b\u4e0d\u61c2\u5c31\u9019\u6a23\u554f AI<\/h2>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u7528\u767d\u8a71\u4e00\u6bb5\u4e00\u6bb5\u89e3\u91cb\u9019\u4efd MLflow \u8a55\u4f30\u8173\u672c\u5728\u8a55\u4ec0\u9ebc\uff1a\u5b83\u7528\u4e86\u54ea\u4e9b scorer\u3001\u6bcf\u500b scorer \u9700\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\u3001\u5224\u6e96\u5beb\u5728\u54ea\u3002\u6211\u8981\u80fd\u5c0d\u5225\u4eba\u8b1b\u6e05\u695a\u3002\u300d<\/p>\n<\/blockquote>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u9019\u500b make_judge \u7684 instructions \u5224\u6e96\u5920\u5177\u9ad4\u55ce\uff1f\u6703\u4e0d\u6703\u592a\u5bec\u9b06\u5c0e\u81f4\u4ec0\u9ebc\u7b54\u6848\u90fd pass\uff1f\u5e6b\u6211\u5217\u51fa\u53ef\u4ee5\u6539\u9032\u7684\u5730\u65b9\u3002\u300d<\/p>\n<\/blockquote>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\">\u300c\u5e6b\u6211\u6aa2\u67e5\u9019\u4efd\u8a55\u4f30\u96c6\u6709\u6c92\u6709\u8cc7\u6599\u6d29\u6f0f\u2014\u2014\u6a19\u6e96\u7b54\u6848\u6709\u6c92\u6709\u4e0d\u5c0f\u5fc3\u51fa\u73fe\u5728\u6703\u9935\u7d66 app \u7684 inputs \u88e1\uff1f\u300d<\/p>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5c0f\u7d50<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n\n<li>LLM \u8f38\u51fa\u6c92\u6709\u552f\u4e00\u6b63\u89e3\uff0c\u6240\u4ee5\u7528 <strong>LLM-as-a-Judge<\/strong> \u6253\u5206\uff0cMLflow \u7528 <code>mlflow.genai.evaluate<\/code> \u628a\u5b83\u6a19\u6e96\u5316<\/li>\n\n\n<li>\u8a55\u4f30\u96c6\u7684\u5730\u57fa\u662f <strong><code>inputs<\/code>\uff08\u9935\u7d66 app\uff09+ <code>expectations<\/code>\uff08\u4f60\u7684\u671f\u671b\uff09<\/strong>\uff1b\u6a19\u6e96\u7b54\u6848\u53ea\u653e expectations<\/li>\n\n\n<li><strong>\u5167\u5efa scorer<\/strong>\uff08Correctness \/ RelevanceToQuery \/ RetrievalGroundedness \/ Guidelines\uff09\u80fd\u5feb\u901f\u4e0a\u624b\uff0c\u5176\u4e2d\u6709\u4e9b\u9700\u8981\u6a19\u6e96\u7b54\u6848\u3001\u6709\u4e9b\u4e0d\u7528<\/li>\n\n\n<li>\u81ea\u8a02 judge \u7528 <strong><code>make_judge<\/code><\/strong>\uff0c\u5224\u6e96\uff08rubric\uff09\u5c31\u662f\u90a3\u6bb5 <code>instructions<\/code>\uff0c\u8d8a\u5177\u9ad4\u8d8a\u53ef\u9760<\/li>\n\n\n<li>\u8b80\u7d50\u679c\u8868\u683c\u8981\u6a6b\u770b\u55ae\u984c\u3001\u76f4\u770b\u8da8\u52e2\uff1b\u6bd4\u8f03\u7248\u672c\u8981\u56fa\u5b9a data \u548c scorers<\/li>\n\n\n<li>\u56db\u5927\u7d05\u65d7\uff1a<strong>judge \u592a\u5bec\u9b06\u3001\u8cc7\u6599\u6d29\u6f0f\u3001\u7403\u54e1\u517c\u88c1\u5224\u3001\u5b8c\u5168\u4e0d\u505a\u4eba\u5de5\u62bd\u67e5<\/strong><\/li>\n\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0b\u4e00\u7bc7\uff08#03\uff09\u6211\u5011\u6703\u628a\u8a55\u4f30\u63a5\u9032\u5be6\u969b\u7684\u8fed\u4ee3\u6d41\u7a0b\uff0c\u770b\u600e\u9ebc\u7528\u8a55\u4f30\u7d50\u679c\u9a45\u52d5 prompt \u512a\u5316\u3002<\/p>\n\n\n\n<div id=\"quiz-94442124\">\n<style>\n#quiz-94442124 .badge {\n        display: inline-block; padding: 4px 10px; border-radius: 12px;\n        font-size: 0.75em; font-weight: bold; margin-left: 8px;\n      }#quiz-94442124 .badge.scenario { background: #ebf8ff; color: #2b6cb0; }#quiz-94442124 .badge.debug { background: #fef3c7; color: #92400e; }#quiz-94442124 .question-adv {\n        background: white; border-radius: 10px; padding: 24px; margin: 20px 0;\n        box-shadow: 0 2px 8px rgba(0, 0, 0, 0.08); border-left: 4px solid #ed8936;\n      }#quiz-94442124 .question-adv h3 { margin-top: 0; color: #2d3748; }#quiz-94442124 .code-block {\n        background: #1e1e1e; color: #d4d4d4; padding: 14px; border-radius: 6px;\n        font-family: \"Menlo\", \"Monaco\", \"Courier New\", monospace; font-size: 0.9em;\n        overflow-x: auto; margin: 12px 0; white-space: pre-wrap;\n      }#quiz-94442124 .options-adv { list-style: none; padding: 0; }#quiz-94442124 .options-adv li {\n        padding: 14px 16px; margin: 10px 0; border: 2px solid #e2e8f0;\n        border-radius: 8px; cursor: pointer; transition: all 0.2s;\n      }#quiz-94442124 .options-adv li:hover:not(.disabled) { border-color: #ed8936; background: #fffaf0; }#quiz-94442124 .options-adv li.disabled { cursor: default; opacity: 0.7; }#quiz-94442124 .options-adv li.show-answer { border-color: #48bb78; background: #f0fff4; }#quiz-94442124 .options-adv li.wrong { border-color: #fc8181; background: #fff5f5; }#quiz-94442124 .feedback-adv { margin-top: 16px; padding: 14px; border-radius: 8px; display: none; line-height: 1.6; }#quiz-94442124 .feedback-adv.correct { background: #f0fff4; color: #276749; display: block; }#quiz-94442124 .feedback-adv.wrong { background: #fff5f5; color: #c53030; display: block; }#quiz-94442124 .feedback-adv \n      .score-adv {\n        text-align: center; font-size: 1.4em; padding: 24px; background: white;\n        border-radius: 10px; margin-top: 24px; box-shadow: 0 2px 8px rgba(0, 0, 0, 0.08);\n      }#quiz-94442124 .intro { background: #ebf8ff; padding: 16px; border-radius: 8px; margin-bottom: 20px; color: #2c5282; }\n<\/style>\n<h1>\u9032\u968e\u6e2c\u9a57\uff1a\u7528 LLM-as-a-Judge \u8a55\u4f30\u4f60\u7684\u6a21\u578b\u8f38\u51fa<\/h1>\n    <div class=\"intro\">\n      <strong>\u6e2c\u9a57\u76ee\u6a19\uff1a<\/strong>\u9a57\u8b49\u4f60\u662f\u5426\u80fd\u5728\u5be6\u969b\u60c5\u5883\u4e2d\u61c9\u7528\u6240\u5b78\u3002<br \/>\n      \u5171 5 \u984c\uff0c\u5305\u542b\u60c5\u5883\u984c\u8207\u932f\u8aa4\u8a3a\u65b7\u984c\u3002\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv1\">\n      <h3>1. \u4f60\u8981\u8a55\u4f30\u4e00\u500b RAG \u5ba2\u670d\u6a5f\u5668\u4eba\uff0c\u91cd\u9ede\u662f\u300c\u7b54\u6848\u6709\u6c92\u6709\u6839\u64da\u6aa2\u7d22\u5230\u7684 context\u3001\u9084\u662f\u5728\u5e7b\u89ba\u300d\uff0c\u4f46\u4f60\u624b\u4e0a\u6c92\u6709\u4eba\u5de5\u6a19\u597d\u7684\u6a19\u6e96\u7b54\u6848\u3002\u4f60\u8a72\u9078\u54ea\u500b scorer\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(1, 'A', 'B')\">A. <code>Correctness()<\/code>\uff0c\u56e0\u70ba\u5b83\u6700\u5168\u9762<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'B', 'B')\">B. <code>RetrievalGroundedness()<\/code>\uff0c\u5b83\u6e2c\u56de\u7b54\u6709\u6c92\u6709\u6839\u64da context\u3001\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'C', 'B')\">C. \u53ea\u80fd\u5148\u82b1\u6642\u9593\u4eba\u5de5\u6a19\u597d expected_response\uff0c\u5426\u5247\u7121\u6cd5\u8a55\u4f30<\/li>\n        <li onclick=\"checkAnswerAdv(1, 'D', 'B')\">D. \u7528 exact match \u6bd4\u5c0d\u5c31\u597d<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv1\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv2\">\n      <h3>2. \u4f60\u60f3\u6bd4\u8f03 prompt v1 \u8207 v2 \u54ea\u500b\u6bd4\u8f03\u597d\uff0c\u6b63\u78ba\u7684\u505a\u6cd5\u662f\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(2, 'A', 'C')\">A. v1 \u7528 correctness\u3001v2 \u7528 relevance\uff0c\u770b\u54ea\u500b\u5206\u6578\u9ad8<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'B', 'C')\">B. \u5169\u500b\u7248\u672c\u5404\u81ea\u6e96\u5099\u6700\u80fd\u51f8\u986f\u81ea\u5df1\u512a\u9ede\u7684\u8a55\u4f30\u96c6<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'C', 'C')\">C. \u7528\u540c\u4e00\u4efd data \u548c\u540c\u4e00\u7d44 scorers\uff0c\u5404\u81ea start_run \u8dd1\u4e00\u6b21\uff0c\u53ea\u63db predict_fn\uff0c\u518d\u5230 UI \u4e26\u6392\u6bd4\u8f03<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'D', 'C')\">D. \u53ea\u8dd1 v2\uff0c\u56e0\u70ba\u5b83\u662f\u65b0\u7684\uff0c\u4e00\u5b9a\u6bd4\u8f03\u597d<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv2\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv3\">\n      <h3>3. \u4f60\u88ab\u6307\u6d3e\u8a55\u4f30\u4e00\u500b\u300c\u628a\u4f7f\u7528\u8005\u8f38\u5165\u7684\u554f\u984c\u6539\u5beb\u5f97\u66f4\u5ba2\u6c23\u300d\u7684 app\uff0c\u91cd\u9ede\u662f\u8a9e\u6c23\u3002\u4f60\u60f3\u7528\u6700\u7701\u4e8b\u3001\u4e0d\u9700\u6a19\u6e96\u7b54\u6848\u7684\u65b9\u5f0f\u8a55\u300c\u6709\u6c92\u6709\u4fdd\u6301\u79ae\u8c8c\u5c08\u696d\u300d\u3002\u6700\u5408\u9069\u7684\u505a\u6cd5\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(3, 'A', 'A')\">A. \u7528 <code>Guidelines(name=..., guidelines=\"\u56de\u7b54\u5fc5\u9808\u79ae\u8c8c\u5c08\u696d\u2026\")<\/code>\uff0c\u7528\u767d\u8a71\u5beb\u4e00\u689d\u898f\u5247\u8b93 judge \u5224<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'B', 'A')\">B. \u7528 <code>Correctness()<\/code> \u914d\u4e00\u4efd\u9010\u5b57\u6a19\u597d\u7684\u6a19\u6e96\u7b54\u6848<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'C', 'A')\">C. \u7528\u5b57\u4e32\u9577\u5ea6\u5224\u65b7\uff0c\u8d8a\u9577\u8d8a\u79ae\u8c8c<\/li>\n        <li onclick=\"checkAnswerAdv(3, 'D', 'A')\">D. \u7121\u6cd5\u7528 MLflow \u8a55\u4f30\u98a8\u683c\uff0c\u53ea\u80fd\u7d14\u4eba\u5de5<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv3\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv4\">\n      <h3>4. \u540c\u4e8b\u7684\u8a55\u4f30\u96c6\u9019\u6a23\u5beb\uff0c\u4e14\u8a55\u4f30\u5206\u6578\u7570\u5e38\u5730\u9ad8\uff0c\u4e0a\u7dda\u5f8c\u537b\u8868\u73fe\u5f88\u5dee\u3002\u554f\u984c\u6700\u53ef\u80fd\u51fa\u5728\u54ea\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">data = [{\n    &#8220;inputs&#8221;: {\n        &#8220;question&#8221;: &#8220;\u9000\u8ca8\u5e7e\u5929\u5167\uff1f&#8221;,\n        &#8220;answer&#8221;: &#8220;7 \u5929&#8221;,\n    },\n    &#8220;expectations&#8221;: {&#8220;expected_response&#8221;: &#8220;7 \u5929&#8221;},\n}]<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(4, 'A', 'B')\">A. expectations \u4e0d\u8a72\u6709 expected_response<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'B', 'B')\">B. \u8cc7\u6599\u6d29\u6f0f\uff1a\u6a19\u6e96\u7b54\u6848 &#8220;answer&#8221; \u88ab\u653e\u9032 inputs \u9935\u7d66 app\uff0c\u7b49\u65bc\u8003\u5377\u5370\u7b54\u6848\uff0c\u5206\u6578\u865b\u9ad8<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'C', 'B')\">C. inputs \u4e0d\u80fd\u6709 question \u6b04\u4f4d<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'D', 'B')\">D. \u7f3a\u5c11 scorers \u53c3\u6578\uff08\u6b64\u6bb5\u672c\u4f86\u5c31\u6c92\u6709\uff09<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv4\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv5\">\n      <h3>5. \u4e00\u4efd\u81ea\u8a02 judge \u5e7e\u4e4e\u6240\u6709\u56de\u7b54\u90fd\u5224 pass\uff0c\u5718\u968a\u537b\u89ba\u5f97\u6a21\u578b\u660e\u660e\u5e38\u51fa\u932f\u3002\u770b\u5230\u4e0b\u9762\u7684 judge \u5b9a\u7fa9\uff0c\u6700\u8a72\u512a\u5148\u4fee\u6b63\u4ec0\u9ebc\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">quality = make_judge(\n    name=&#8221;quality&#8221;,\n    instructions=&#8221;\u5224\u65b7\u9019\u500b\u56de\u7b54\u597d\u4e0d\u597d\u3002{{ outputs }}&#8221;,\n    model=&#8221;openai:\/gpt-4o&#8221;,\n)<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(5, 'A', 'C')\">A. \u628a model \u63db\u6210\u66f4\u4fbf\u5b9c\u7684\u6a21\u578b<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'B', 'C')\">B. \u628a name \u6539\u5f97\u66f4\u9577\u66f4\u63cf\u8ff0\u6027<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'C', 'C')\">C. instructions \u5224\u6e96\u592a\u5bec\u9b06\uff08\u300c\u597d\u4e0d\u597d\u300d\u7121\u53ef\u6aa2\u67e5\u6a19\u6e96\uff09\uff0c\u6539\u6210\u5217\u51fa\u660e\u78ba\u53ef\u6aa2\u67e5\u7684\u689d\u4ef6\u4e26\u8981\u6c42 pass\/fail \u52a0\u7406\u7531<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'D', 'C')\">D. \u79fb\u9664 {{ outputs }} \u4f54\u4f4d\u7b26<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv5\"><\/div>\n    <\/div>\n\n    <div class=\"score-adv\" id=\"scoreAdv\" style=\"display:none;\"><\/div>\n<script>\n\n      let answeredAdv = 0;\n      let correctAdv = 0;\n      const totalAdv = 5;\n      const explanationsAdv = {\n        1: \"RetrievalGroundedness \u5c08\u9580\u6e2c\u300c\u56de\u7b54\u6709\u6c92\u6709\u6839\u64da\u6aa2\u7d22\u5230\u7684 context\u3001\u9084\u662f\u5728\u5e7b\u89ba\u300d\uff0c\u800c\u4e14\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\uff0c\u6b63\u597d\u7b26\u5408\u300c\u6c92\u6709\u4eba\u5de5\u6a19\u8a3b\u300d\u7684\u9650\u5236\u3002Correctness \u9700\u8981 expected_response\uff1bexact match \u5c0d LLM \u8f38\u51fa\u5e7e\u4e4e\u7121\u7528\u3002\",\n        2: \"\u6bd4\u8f03\u7684\u6838\u5fc3\u662f\u8b8a\u56e0\u55ae\u4e00\uff1a\u56fa\u5b9a\u540c\u4e00\u4efd data \u8207\u540c\u4e00\u7d44 scorers\uff0c\u53ea\u63db predict_fn\uff0c\u5404\u81ea start_run \u7522\u751f\u5169\u500b\u53ef\u6bd4\u7684 run\u3002\u7528\u4e0d\u540c\u8a55\u4f30\u96c6\u6216\u4e0d\u540c scorer \u6703\u8b93\u5206\u6578\u4e0d\u53ef\u6bd4\uff1b\u53ea\u8dd1\u65b0\u7248\u7121\u5f9e\u6bd4\u8f03\u3002\",\n        3: \"Guidelines \u8b93\u4f60\u7528\u81ea\u7136\u8a9e\u8a00\u5beb\u4e00\u689d\u898f\u5247\uff08\u5982\u300c\u5fc5\u9808\u79ae\u8c8c\u5c08\u696d\u300d\uff09\uff0cjudge \u8b80\u6bcf\u500b\u56de\u7b54\u5224 pass\/fail\uff0c\u4e0d\u9700\u8981\u6a19\u6e96\u7b54\u6848\uff0c\u6700\u9069\u5408\u8a55\u300c\u98a8\u683c\/\u8a9e\u6c23\u6709\u6c92\u6709\u5b88\u898f\u77e9\u300d\u3002Correctness \u9700\u8981\u9010\u5b57\u6a19\u6e96\u7b54\u6848\uff0c\u6210\u672c\u9ad8\u53c8\u4e0d\u9069\u5408\u8a55\u98a8\u683c\u3002\",\n        4: \"inputs \u662f\u6703\u88ab\u9935\u9032 app \u7684\u6771\u897f\u3002\u628a\u6a19\u6e96\u7b54\u6848 answer \u593e\u5728 inputs \u88e1\uff0c\u7b49\u65bc\u8003\u8a66\u628a\u7b54\u6848\u5370\u5728\u8003\u5377\u4e0a\uff0capp \u76f4\u63a5\u6284\uff0c\u5206\u6578\u865b\u9ad8\uff0c\u4e0a\u7dda\u5c31\u73fe\u5f62\u3002\u6a19\u6e96\u7b54\u6848\u53ea\u80fd\u653e\u5728 expectations\u3002\",\n        5: \"\u300c\u5224\u65b7\u597d\u4e0d\u597d\u300d\u6c92\u6709\u53ef\u6aa2\u67e5\u7684\u6a19\u6e96\uff0cjudge \u53ea\u80fd\u6191\u611f\u89ba\u3001\u5e7e\u4e4e\u5168 pass\uff0c\u7b49\u65bc\u6c92\u8a55\u3002\u4fee\u6b63\u65b9\u5411\u662f\u628a instructions \u5beb\u5177\u9ad4\uff1a\u5217\u51fa\u660e\u78ba\u53ef\u6aa2\u67e5\u7684\u689d\u4ef6\uff08\u8981\u542b\u54ea\u4e9b\u8cc7\u8a0a\u3001\u8a9e\u6c23\u3001\u9577\u5ea6\u2026\uff09\u4e26\u8981\u6c42\u8f38\u51fa pass\/fail \u52a0\u7406\u7531\u3002\u63db\u4fbf\u5b9c\u6a21\u578b\u6216\u6539\u540d\u5b57\u90fd\u7121\u52a9\u65bc\u5224\u6e96\u54c1\u8cea\u3002\",\n      };\n\n      function checkAnswerAdv(qNum, selected, correctAns) {\n        const question = document.getElementById(\"qAdv\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n        question.classList.add(\"answered\");\n        answeredAdv++;\n        const options = question.querySelectorAll(\".options-adv li\");\n        options.forEach((opt) => {\n          opt.classList.add(\"disabled\");\n          if (opt.textContent.startsWith(correctAns + \".\")) {\n            opt.classList.add(\"show-answer\");\n          }\n        });\n        const feedback = document.getElementById(\"fbAdv\" + qNum);\n        if (selected === correctAns) {\n          correctAdv++;\n          feedback.className = \"feedback-adv correct\";\n          feedback.innerHTML = \"\u2713 \u6b63\u78ba\uff01\" + explanationsAdv[qNum];\n        } else {\n          feedback.className = \"feedback-adv wrong\";\n          feedback.innerHTML = \"\u2717 \u932f\u8aa4\u3002\u6b63\u78ba\u7b54\u6848\u662f \" + correctAns + \"\u3002\" + explanationsAdv[qNum];\n          options.forEach((opt) => {\n            if (opt.textContent.startsWith(selected + \".\")) {\n              opt.classList.add(\"wrong\");\n            }\n          });\n        }\n        if (answeredAdv === totalAdv) {\n          const scoreDiv = document.getElementById(\"scoreAdv\");\n          scoreDiv.style.display = \"block\";\n          let msg = \"\";\n          if (correctAdv === totalAdv) msg = \"\ud83c\udfaf \u5b8c\u7f8e\uff01\u4f60\u5df2\u7d93\u6e96\u5099\u597d\u5be6\u6230\u4e86\uff01\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.8)) msg = \"\ud83d\udcaa \u5f88\u68d2\uff01\u518d\u8907\u7fd2\u4e00\u4e0b\u932f\u7684\u984c\u76ee\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.6)) msg = \"\ud83d\udcd6 \u5efa\u8b70\u91cd\u8b80\u76f8\u95dc\u6bb5\u843d\u5f8c\u518d\u8a66\u4e00\u6b21\";\n          else msg = \"\ud83d\udd04 \u5efa\u8b70\u91cd\u65b0\u95b1\u8b80\u6587\u4ef6\uff0c\u7279\u5225\u6ce8\u610f\u64cd\u4f5c\u6d41\u7a0b\";\n          scoreDiv.innerHTML = \"\u6e2c\u9a57\u5b8c\u6210\uff01\u5f97\u5206\uff1a\" + correctAdv + \" \/ \" + totalAdv + \"<br>\" + msg;\n        }\n      }\n    \n<\/script>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u6e2c\u9a57\uff1a\u7528 LLM-as-a-Judge \u8a55\u4f30\u4f60\u7684\u6a21\u578b\u8f38\u51fa \u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50\u679c 1. \u6587\u7ae0\u6307\u51fa [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rolpb_likes":0,"footnotes":""},"categories":[223],"tags":[881,880,883,392,882],"class_list":["post-848","post","type-post","status-publish","format-standard","hentry","category-mlflow","tag-genai","tag-llm-as-a-judge-2","tag-llmops","tag-mlflow","tag-mlflow-genai-evaluate"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":7}},"_links":{"self":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/848","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/comments?post=848"}],"version-history":[{"count":0,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/848\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/media?parent=848"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/categories?post=848"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/tags?post=848"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}