{"id":443,"date":"2026-01-27T18:10:41","date_gmt":"2026-01-27T10:10:41","guid":{"rendered":"https:\/\/blog.itsmygo.uk\/2026\/01\/27\/%e3%80%90mlflow-genai-%e6%95%99%e5%ad%b8%e3%80%9104-evaluation%ef%bc%9a%e7%94%a8-llm-%e8%a9%95%e4%bc%b0-llm-%e8%bc%b8%e5%87%ba%e5%93%81%e8%b3%aa\/"},"modified":"2026-01-27T18:10:41","modified_gmt":"2026-01-27T10:10:41","slug":"%e3%80%90mlflow-genai-%e6%95%99%e5%ad%b8%e3%80%9104-evaluation%ef%bc%9a%e7%94%a8-llm-%e8%a9%95%e4%bc%b0-llm-%e8%bc%b8%e5%87%ba%e5%93%81%e8%b3%aa","status":"publish","type":"post","link":"https:\/\/blog.itsmygo.uk\/wordpress\/2026\/01\/27\/%e3%80%90mlflow-genai-%e6%95%99%e5%ad%b8%e3%80%9104-evaluation%ef%bc%9a%e7%94%a8-llm-%e8%a9%95%e4%bc%b0-llm-%e8%bc%b8%e5%87%ba%e5%93%81%e8%b3%aa\/","title":{"rendered":"\u3010MLflow GenAI \u6559\u5b78\u3011#04 Evaluation\uff1a\u7528 LLM \u8a55\u4f30 LLM \u8f38\u51fa\u54c1\u8cea"},"content":{"rendered":"\n<div id=\"quiz-334e6c14\">\n<style>\n#quiz-334e6c14 .question {\n        background: white;\n        border-radius: 8px;\n        padding: 20px;\n        margin: 20px 0;\n        box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1);\n      }#quiz-334e6c14 .question h3 {\n        margin-top: 0;\n        color: #333;\n      }#quiz-334e6c14 .options {\n        list-style: none;\n        padding: 0;\n      }#quiz-334e6c14 .options li {\n        padding: 12px 15px;\n        margin: 8px 0;\n        border: 2px solid #ddd;\n        border-radius: 6px;\n        cursor: pointer;\n        transition: all 0.2s;\n      }#quiz-334e6c14 .options li:hover:not(.disabled) {\n        border-color: #4caf50;\n        background: #f0fff0;\n      }#quiz-334e6c14 .options li.correct {\n        border-color: #4caf50;\n        background: #e8f5e9;\n      }#quiz-334e6c14 .options li.wrong {\n        border-color: #f44336;\n        background: #ffebee;\n      }#quiz-334e6c14 .options li.disabled {\n        cursor: default;\n        opacity: 0.7;\n      }#quiz-334e6c14 .options li.show-answer {\n        border-color: #4caf50;\n        background: #e8f5e9;\n      }#quiz-334e6c14 .feedback {\n        margin-top: 15px;\n        padding: 12px;\n        border-radius: 6px;\n        display: none;\n      }#quiz-334e6c14 .feedback.correct {\n        background: #e8f5e9;\n        color: #2e7d32;\n        display: block;\n      }#quiz-334e6c14 .feedback.wrong {\n        background: #ffebee;\n        color: #c62828;\n        display: block;\n      }#quiz-334e6c14 .score {\n        text-align: center;\n        font-size: 1.5em;\n        padding: 20px;\n        background: white;\n        border-radius: 8px;\n        margin-top: 20px;\n      }#quiz-334e6c14 .code-block {\n        background: #1e1e1e;\n        color: #d4d4d4;\n        padding: 14px;\n        border-radius: 6px;\n        font-family: \"Monaco\", \"Consolas\", \"Courier New\", monospace;\n        font-size: 0.9em;\n        overflow-x: auto;\n        margin: 12px 0;\n        white-space: pre-wrap;\n      }\n<\/style>\n<h1>\u6e2c\u9a57\uff1aMLflow GenAI \u6559\u5b78 &#8211; Evaluation<\/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. MLflow Evaluation \u63a1\u7528\u7684\u6838\u5fc3\u8a55\u4f30\u65b9\u6cd5\u7a31\u70ba\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(1, 'A', 'C')\">A. Human-in-the-Loop<\/li>\n        <li onclick=\"checkAnswer(1, 'B', 'C')\">B. A\/B Testing<\/li>\n        <li onclick=\"checkAnswer(1, 'C', 'C')\">C. LLM-as-a-Judge<\/li>\n        <li onclick=\"checkAnswer(1, 'D', 'C')\">D. Reinforcement Learning<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb1\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q2\">\n      <h3>2. \u5728 MLflow \u8a55\u4f30\u6846\u67b6\u4e2d\uff0c\u5b9a\u7fa9\u8a55\u4f30\u6a19\u6e96\u7684\u6838\u5fc3\u5143\u4ef6\u7a31\u70ba\u4ec0\u9ebc\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(2, 'A', 'B')\">A. Evaluator<\/li>\n        <li onclick=\"checkAnswer(2, 'B', 'B')\">B. Scorer<\/li>\n        <li onclick=\"checkAnswer(2, 'C', 'B')\">C. Metric<\/li>\n        <li onclick=\"checkAnswer(2, 'D', 'B')\">D. Judge<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb2\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q3\">\n      <h3>3. \u4e0b\u5217\u54ea\u500b Scorer \u4e0d\u9700\u8981 Ground Truth\uff08\u9810\u671f\u7b54\u6848\uff09\u5c31\u80fd\u904b\u4f5c\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(3, 'A', 'D')\">A. Correctness<\/li>\n        <li onclick=\"checkAnswer(3, 'B', 'D')\">B. Equivalence<\/li>\n        <li onclick=\"checkAnswer(3, 'C', 'D')\">C. Guidelines<\/li>\n        <li onclick=\"checkAnswer(3, 'D', 'D')\">D. RelevanceToQuery<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb3\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q4\">\n      <h3>4. \u4f7f\u7528 <code>mlflow.genai.evaluate()<\/code> \u6642\uff0c\u5fc5\u9808\u63d0\u4f9b\u54ea\u4e09\u500b\u6838\u5fc3\u5143\u7d20\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(4, 'A', 'B')\">A. Model\u3001Prompt\u3001Temperature<\/li>\n        <li onclick=\"checkAnswer(4, 'B', 'B')\">B. Dataset\u3001Predict Function\u3001Scorers<\/li>\n        <li onclick=\"checkAnswer(4, 'C', 'B')\">C. Input\u3001Output\u3001Metrics<\/li>\n        <li onclick=\"checkAnswer(4, 'D', 'B')\">D. Experiment\u3001Run\u3001Artifacts<\/li>\n      <\/ul>\n      <div class=\"feedback\" id=\"fb4\"><\/div>\n    <\/div>\n\n    <div class=\"question\" id=\"q5\">\n      <h3>5. \u6839\u64da MLflow \u6700\u4f73\u5be6\u8e10\uff0c\u8a55\u5206\u65b9\u5f0f\u61c9\u8a72\u512a\u5148\u4f7f\u7528\u54ea\u7a2e\u985e\u578b\uff1f<\/h3>\n      <ul class=\"options\">\n        <li onclick=\"checkAnswer(5, 'A', 'C')\">A. 1-10 \u7684\u6578\u503c\u8a55\u5206<\/li>\n        <li onclick=\"checkAnswer(5, 'B', 'C')\">B. \u767e\u5206\u6bd4\u8a55\u5206<\/li>\n        <li onclick=\"checkAnswer(5, 'C', 'C')\">C. \u4e8c\u5143\uff08pass\/fail\uff09\u6216\u5c11\u91cf\u985e\u5225\u7684\u8a55\u5206<\/li>\n        <li onclick=\"checkAnswer(5, 'D', 'C')\">D. \u9023\u7e8c\u6578\u503c\u8a55\u5206\uff080.0-1.0\uff09<\/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: \"MLflow \u63a1\u7528\u7684\u6838\u5fc3\u65b9\u6cd5\u53eb\u505a <strong>LLM-as-a-Judge<\/strong>\uff0c\u7528\u4e00\u500b LLM \u4f86\u8a55\u4f30\u53e6\u4e00\u500b LLM \u7684\u8f38\u51fa\u3002\u7814\u7a76\u986f\u793a\uff0c\u7576\u8a55\u4f30\u6a19\u6e96\u660e\u78ba\u5b9a\u7fa9\u6642\uff0cLLM \u5224\u65b7\u7684\u4e00\u81f4\u6027\u76f8\u7576\u9ad8\u3002\",\n        2: \"\u5728 MLflow \u7684\u8a55\u4f30\u6846\u67b6\u4e2d\uff0c<strong>Scorer<\/strong> \u662f\u5b9a\u7fa9\u8a55\u4f30\u6a19\u6e96\u7684\u6838\u5fc3\u5143\u4ef6\u3002\u4f60\u53ef\u4ee5\u628a\u5b83\u60f3\u6210\u4e00\u500b\u8a55\u5206\u5668\uff0c\u8f38\u5165\u662f\u6a21\u578b\u7684\u8f38\u51fa\uff0c\u8f38\u51fa\u662f\u54c1\u8cea\u5206\u6578\u3002\",\n        3: \"<code>RelevanceToQuery<\/code> \u4e0d\u9700\u8981 Ground Truth \u5c31\u80fd\u904b\u4f5c\uff0c\u5b83\u8a55\u4f30\u7684\u662f\u56de\u7b54\u662f\u5426\u91dd\u5c0d\u554f\u984c\u3002\u800c <code>Correctness<\/code>\u3001<code>Equivalence<\/code> \u548c <code>Guidelines<\/code> \u90fd\u9700\u8981\u9810\u671f\u7b54\u6848\u6216\u53c3\u8003\u6a19\u6e96\u3002\",\n        4: \"\u8a55\u4f30\u6d41\u7a0b\u6709\u4e09\u500b\u6838\u5fc3\u5143\u7d20\uff1a<strong>Dataset<\/strong>\uff08\u8a55\u4f30\u8cc7\u6599\u96c6\uff09\u3001<strong>Predict Function<\/strong>\uff08\u4f60\u7684\u6a21\u578b\u6216\u61c9\u7528\uff09\u3001<strong>Scorers<\/strong>\uff08\u8a55\u4f30\u6a19\u6e96\uff09\u3002\",\n        5: \"MLflow \u5efa\u8b70\u4f7f\u7528<strong>\u4e8c\u5143\uff08pass\/fail\uff09\u6216\u5c11\u91cf\u985e\u5225\u7684\u8a55\u5206<\/strong>\uff0c\u800c\u975e 1-5 \u7684\u6578\u503c\u8a55\u5206\u3002\u539f\u56e0\u662f LLM \u7522\u751f\u7684\u5224\u65b7\u66f4\u4e00\u81f4\u3001\u66f4\u5bb9\u6613\u8a2d\u5b9a\u95be\u503c\u3001\u6c7a\u7b56\u66f4\u660e\u78ba\u3002\"\n      };\n\n      function checkAnswer(qNum, selected, correctAns) {\n        const question = document.getElementById(\"q\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n\n        question.classList.add(\"answered\");\n        answered++;\n\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\n        const feedback = document.getElementById(\"fb\" + qNum);\n        if (selected === correctAns) {\n          correct++;\n          feedback.className = \"feedback correct\";\n          feedback.innerHTML = \"&#10003; \u6b63\u78ba\uff01\" + explanations[qNum];\n        } else {\n          feedback.className = \"feedback wrong\";\n          feedback.innerHTML =\n            \"&#10007; \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\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\" +\n            correct +\n            \" \/ \" +\n            total +\n            (correct === total\n              ? \" \u5168\u5c0d\uff01\"\n              : correct >= Math.ceil(total * 0.6)\n                ? \" \u4e0d\u932f\uff01\"\n                : \" \u518d\u8907\u7fd2\u4e00\u4e0b\u5427\");\n        }\n      }\n    \n<\/script>\n<\/div>\n\n\n\n<h2 class=\"wp-block-heading\">\u70ba\u4ec0\u9ebc\u9700\u8981\u81ea\u52d5\u5316\u8a55\u4f30\uff1f<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u4f60\u958b\u767c GenAI \u61c9\u7528\u6642\uff0c\u4e00\u5b9a\u6703\u9047\u5230\u9019\u500b\u554f\u984c\uff1a<strong>\u600e\u9ebc\u77e5\u9053\u6a21\u578b\u8f38\u51fa\u7684\u54c1\u8cea\u597d\u4e0d\u597d\uff1f<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u50b3\u7d71\u7684\u505a\u6cd5\u662f\u4eba\u5de5\u6e2c\u8a66\u2014\u2014\u5beb\u5e7e\u500b\u554f\u984c\u3001\u770b\u770b\u56de\u7b54\u3001\u6191\u611f\u89ba\u5224\u65b7\u3002\u4f46\u9019\u7a2e\u65b9\u5f0f\u6709\u5e7e\u500b\u81f4\u547d\u7f3a\u9677\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u7121\u6cd5\u898f\u6a21\u5316<\/strong>\uff1a\u6e2c 10 \u500b\u6848\u4f8b\u9084\u884c\uff0c\u6e2c 1000 \u500b\u5c31\u5d29\u6f70<\/li>\n<li><strong>\u54c1\u8cea\u96e3\u4ee5\u91cf\u5316<\/strong>\uff1a\u300c\u611f\u89ba\u4e0d\u932f\u300d\u7121\u6cd5\u8ffd\u8e64\u6539\u5584\u5e45\u5ea6<\/li>\n<li><strong>\u7121\u6cd5\u6301\u7e8c\u76e3\u63a7<\/strong>\uff1aProduction \u74b0\u5883\u7684\u54c1\u8cea\u9000\u5316\u5f88\u96e3\u767c\u73fe<\/li>\n<li><strong>\u96e3\u4ee5\u6bd4\u8f03\u7248\u672c<\/strong>\uff1a\u65b0 prompt \u771f\u7684\u6bd4\u820a\u7684\u597d\u55ce\uff1f<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow Evaluation \u89e3\u6c7a\u7684\u6b63\u662f\u9019\u4e9b\u554f\u984c\u3002\u5b83\u8b93\u4f60\u7528\u7a0b\u5f0f\u5316\u7684\u65b9\u5f0f\u5b9a\u7fa9\u54c1\u8cea\u6a19\u6e96\uff0c\u7136\u5f8c\u81ea\u52d5\u8a55\u4f30\u6bcf\u4e00\u6b21\u8f38\u51fa\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">LLM-as-a-Judge \u539f\u7406<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u63a1\u7528\u7684\u6838\u5fc3\u65b9\u6cd5\u53eb\u505a <strong>LLM-as-a-Judge<\/strong>\u2014\u2014\u7528\u4e00\u500b LLM \u4f86\u8a55\u4f30\u53e6\u4e00\u500b LLM \u7684\u8f38\u51fa\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u807d\u8d77\u4f86\u6709\u9ede\u5947\u602a\uff1a\u7528 AI \u8a55\u4f30 AI\uff0c\u53ef\u9760\u55ce\uff1f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7814\u7a76\u986f\u793a\uff0c\u7576\u8a55\u4f30\u6a19\u6e96\u660e\u78ba\u5b9a\u7fa9\u6642\uff0cLLM \u5224\u65b7\u7684\u4e00\u81f4\u6027\u5176\u5be6\u76f8\u7576\u9ad8\u3002\u7279\u5225\u662f\u5c0d\u65bc\u4ee5\u4e0b\u4efb\u52d9\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u56de\u7b54\u662f\u5426\u76f8\u95dc<\/li>\n<li>\u5167\u5bb9\u662f\u5426\u6709\u5bb3<\/li>\n<li>\u662f\u5426\u7b26\u5408\u7279\u5b9a\u98a8\u683c\u6307\u5357<\/li>\n<li>\u4e8b\u5be6\u662f\u5426\u6b63\u78ba\uff08\u9700\u8981\u53c3\u8003\u8cc7\u6599\uff09<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u9810\u8a2d\u4f7f\u7528 GPT-4o-mini \u4f5c\u70ba Judge \u6a21\u578b\uff0c\u4f46\u4f60\u53ef\u4ee5\u63db\u6210\u5176\u4ed6\u6a21\u578b\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-1\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.scorers import Correctness\n\n<span class=\"hljs-comment\"># \u4f7f\u7528\u4e0d\u540c\u7684 Judge \u6a21\u578b<\/span>\nCorrectness(model=<span class=\"hljs-string\">\"anthropic:\/claude-4-opus\"<\/span>)\nCorrectness(model=<span class=\"hljs-string\">\"openai:\/gpt-4o\"<\/span>)\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-1\"><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<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">Scorer \u662f\u4ec0\u9ebc\uff1f<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728 MLflow \u7684\u8a55\u4f30\u6846\u67b6\u4e2d\uff0c<strong>Scorer<\/strong> \u662f\u5b9a\u7fa9\u8a55\u4f30\u6a19\u6e96\u7684\u6838\u5fc3\u5143\u4ef6\u3002\u4f60\u53ef\u4ee5\u628a\u5b83\u60f3\u6210\u4e00\u500b\u8a55\u5206\u5668\uff0c\u8f38\u5165\u662f\u6a21\u578b\u7684\u8f38\u51fa\uff0c\u8f38\u51fa\u662f\u54c1\u8cea\u5206\u6578\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u63d0\u4f9b\u4e94\u7a2e Scorer\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u985e\u578b<\/th>\n<th>\u8aaa\u660e<\/th>\n<th>\u9069\u7528\u5834\u666f<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>Predefined Scorers<\/strong><\/td>\n<td>\u5167\u5efa\u7684\u8a55\u4f30\u5668<\/td>\n<td>\u5feb\u901f\u958b\u59cb<\/td>\n<\/tr>\n<tr>\n<td><strong>Guidelines-based<\/strong><\/td>\n<td>\u7528\u81ea\u7136\u8a9e\u8a00\u63cf\u8ff0 pass\/fail \u689d\u4ef6<\/td>\n<td>\u5408\u898f\u6aa2\u67e5\u3001\u98a8\u683c\u9a57\u8b49<\/td>\n<\/tr>\n<tr>\n<td><strong>Template-based<\/strong><\/td>\n<td>\u7528\u6a21\u677f\u8b8a\u6578\u5b9a\u7fa9\u8a55\u4f30\u908f\u8f2f<\/td>\n<td>\u81ea\u8a02\u8907\u96dc\u8a55\u4f30<\/td>\n<\/tr>\n<tr>\n<td><strong>Agent-as-a-Judge<\/strong><\/td>\n<td>\u5206\u6790\u5b8c\u6574 Trace<\/td>\n<td>\u591a\u6b65\u9a5f\u6d41\u7a0b\u8a55\u4f30<\/td>\n<\/tr>\n<tr>\n<td><strong>Code-based<\/strong><\/td>\n<td>\u7d14 Python \u908f\u8f2f<\/td>\n<td>\u78ba\u5b9a\u6027\u898f\u5247\u6aa2\u67e5<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5167\u5efa\u8a55\u4f30\u6307\u6a19<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u63d0\u4f9b\u591a\u500b\u9810\u5b9a\u7fa9\u7684 Scorer\uff0c\u8b93\u4f60\u4e0d\u7528\u5f9e\u96f6\u958b\u59cb\uff1a<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u55ae\u8f2a\u5c0d\u8a71\u8a55\u4f30<\/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-2\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.scorers import (\n    Correctness,          <span class=\"hljs-comment\"># \u7b54\u6848\u6b63\u78ba\u6027\uff08\u9700\u8981 ground truth\uff09<\/span>\n    RelevanceToQuery,     <span class=\"hljs-comment\"># \u56de\u7b54\u662f\u5426\u5207\u984c<\/span>\n    Completeness,         <span class=\"hljs-comment\"># \u662f\u5426\u5b8c\u6574\u56de\u7b54\u6240\u6709\u554f\u984c<\/span>\n    Fluency,              <span class=\"hljs-comment\"># \u8a9e\u6cd5\u6d41\u66a2\u5ea6<\/span>\n    Safety,               <span class=\"hljs-comment\"># \u662f\u5426\u5305\u542b\u6709\u5bb3\u5167\u5bb9<\/span>\n    RetrievalGroundedness, <span class=\"hljs-comment\"># \u56de\u7b54\u662f\u5426\u6709\u6839\u64da\uff08RAG \u61c9\u7528\uff09<\/span>\n    RetrievalRelevance,   <span class=\"hljs-comment\"># \u6aa2\u7d22\u7d50\u679c\u662f\u5426\u76f8\u95dc<\/span>\n    Guidelines,           <span class=\"hljs-comment\"># \u81ea\u8a02\u6307\u5357\u6aa2\u67e5<\/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<h3 class=\"wp-block-heading\">\u9700\u8981 Ground Truth \u7684 Scorer<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>Scorer<\/th>\n<th>\u8aaa\u660e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>Correctness<\/code><\/td>\n<td>\u56de\u7b54\u662f\u5426\u8207\u9810\u671f\u7b54\u6848\u4e00\u81f4<\/td>\n<\/tr>\n<tr>\n<td><code>Equivalence<\/code><\/td>\n<td>\u56de\u7b54\u662f\u5426\u7b49\u50f9\u65bc\u9810\u671f\u8f38\u51fa<\/td>\n<\/tr>\n<tr>\n<td><code>Guidelines<\/code><\/td>\n<td>\u662f\u5426\u7b26\u5408\u7279\u5b9a\u6307\u5357<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e0d\u9700\u8981 Ground Truth \u7684 Scorer<\/h3>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>Scorer<\/th>\n<th>\u8aaa\u660e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>RelevanceToQuery<\/code><\/td>\n<td>\u56de\u7b54\u662f\u5426\u91dd\u5c0d\u554f\u984c<\/td>\n<\/tr>\n<tr>\n<td><code>Completeness<\/code><\/td>\n<td>\u662f\u5426\u5b8c\u6574\u56de\u7b54<\/td>\n<\/tr>\n<tr>\n<td><code>Fluency<\/code><\/td>\n<td>\u8a9e\u6cd5\u662f\u5426\u6d41\u66a2<\/td>\n<\/tr>\n<tr>\n<td><code>Safety<\/code><\/td>\n<td>\u662f\u5426\u5b89\u5168\u7121\u5bb3<\/td>\n<\/tr>\n<tr>\n<td><code>RetrievalGroundedness<\/code><\/td>\n<td>\u56de\u7b54\u662f\u5426\u6709\u6aa2\u7d22\u4f9d\u64da<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">mlflow.genai.evaluate() \u57fa\u672c\u7528\u6cd5<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u5728\u4f86\u770b\u5be6\u969b\u7684\u7a0b\u5f0f\u78bc\u3002\u8a55\u4f30\u6d41\u7a0b\u6709\u4e09\u500b\u6838\u5fc3\u5143\u7d20\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Dataset<\/strong>\uff1a\u8a55\u4f30\u8cc7\u6599\u96c6<\/li>\n<li><strong>Predict Function<\/strong>\uff1a\u4f60\u7684\u6a21\u578b\u6216\u61c9\u7528<\/li>\n<li><strong>Scorers<\/strong>\uff1a\u8a55\u4f30\u6a19\u6e96<\/li>\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">\u6700\u5c0f\u7bc4\u4f8b<\/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-3\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">import mlflow\nfrom mlflow.genai.scorers import Correctness, RelevanceToQuery\n\n<span class=\"hljs-comment\"># 1. \u6e96\u5099\u8a55\u4f30\u8cc7\u6599\u96c6<\/span>\neval_dataset = &#91;\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"MLflow \u53ef\u4ee5\u7ba1\u7406 prompt \u55ce\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {<span class=\"hljs-string\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"\u53ef\u4ee5\uff0cMLflow \u63d0\u4f9b Prompt Registry \u529f\u80fd\"<\/span>},\n    },\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u4ec0\u9ebc\u662f Tracing\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {<span class=\"hljs-string\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"Tracing \u662f\u8ffd\u8e64 LLM \u61c9\u7528\u57f7\u884c\u6d41\u7a0b\u7684\u529f\u80fd\"<\/span>},\n    },\n]\n\n<span class=\"hljs-comment\"># 2. \u5b9a\u7fa9\u4f60\u7684\u61c9\u7528\uff08\u9019\u88e1\u7528\u7c21\u55ae\u51fd\u6578\u793a\u7bc4\uff09<\/span>\ndef my_app(question: str) -&gt; str:\n    <span class=\"hljs-comment\"># \u5be6\u969b\u4e0a\u9019\u88e1\u6703\u547c\u53eb LLM<\/span>\n    <span class=\"hljs-keyword\">return<\/span> <span class=\"hljs-string\">\"MLflow \u63d0\u4f9b\u5b8c\u6574\u7684 prompt \u7ba1\u7406\u529f\u80fd\"<\/span>\n\n<span class=\"hljs-comment\"># 3. \u57f7\u884c\u8a55\u4f30<\/span>\nresults = mlflow.genai.evaluate(\n    data=eval_dataset,\n    predict_fn=my_app,\n    scorers=&#91;Correctness(), RelevanceToQuery()],\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<h3 class=\"wp-block-heading\">\u8cc7\u6599\u96c6\u7d50\u69cb<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u6bcf\u7b46\u8cc7\u6599\u5305\u542b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><code>inputs<\/code>\uff1a\u50b3\u7d66\u6a21\u578b\u7684\u8f38\u5165\uff08dict \u683c\u5f0f\uff09<\/li>\n<li><code>expectations<\/code>\uff1a\u9810\u671f\u7d50\u679c\uff08\u7528\u65bc\u9700\u8981 ground truth \u7684 Scorer\uff09<\/li>\n<li><code>tags<\/code>\uff1a\u53ef\u9078\u7684\u5206\u985e\u6a19\u7c64<\/li>\n<\/ul>\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=\"JSON \/ JSON with Comments\" data-shcb-language-slug=\"json\"><span><code class=\"hljs language-json\">{\n    <span class=\"hljs-attr\">\"inputs\"<\/span>: {<span class=\"hljs-attr\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u5982\u4f55\u4f7f\u7528 Docker\uff1f\"<\/span>},\n    <span class=\"hljs-attr\">\"expectations\"<\/span>: {\n        <span class=\"hljs-attr\">\"expected_response\"<\/span>: <span class=\"hljs-string\">\"...\"<\/span>,\n        <span class=\"hljs-attr\">\"key_concepts\"<\/span>: &#91;<span class=\"hljs-string\">\"container\"<\/span>, <span class=\"hljs-string\">\"image\"<\/span>, <span class=\"hljs-string\">\"dockerfile\"<\/span>]\n    },\n    <span class=\"hljs-attr\">\"tags\"<\/span>: {<span class=\"hljs-attr\">\"topic\"<\/span>: <span class=\"hljs-string\">\"devops\"<\/span>, <span class=\"hljs-attr\">\"difficulty\"<\/span>: <span class=\"hljs-string\">\"beginner\"<\/span>},\n}\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-4\"><span class=\"shcb-language__label\">Code language:<\/span> <span class=\"shcb-language__name\">JSON \/ JSON with Comments<\/span> <span class=\"shcb-language__paren\">(<\/span><span class=\"shcb-language__slug\">json<\/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\">\u4f7f\u7528 Guidelines Scorer<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u5167\u5efa Scorer \u7121\u6cd5\u6eff\u8db3\u9700\u6c42\u6642\uff0c<code>Guidelines<\/code> \u662f\u6700\u7c21\u55ae\u7684\u81ea\u8a02\u65b9\u5f0f\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\">from mlflow.genai.scorers import Guidelines\n\n<span class=\"hljs-comment\"># \u5b9a\u7fa9\u4f60\u7684\u8a55\u4f30\u6307\u5357<\/span>\nprofessional_tone = Guidelines(\n    name=<span class=\"hljs-string\">\"professional_tone\"<\/span>,\n    guidelines=<span class=\"hljs-string\">\"\u56de\u7b54\u5fc5\u9808\u4fdd\u6301\u5c08\u696d\u8a9e\u8abf\uff0c\u4e0d\u4f7f\u7528\u53e3\u8a9e\u6216\u8868\u60c5\u7b26\u865f\"<\/span>\n)\n\nno_code_in_response = Guidelines(\n    name=<span class=\"hljs-string\">\"no_code\"<\/span>,\n    guidelines=<span class=\"hljs-string\">\"\u56de\u7b54\u4e2d\u4e0d\u61c9\u5305\u542b\u7a0b\u5f0f\u78bc\u7bc4\u4f8b\"<\/span>\n)\n\n<span class=\"hljs-comment\"># \u4f7f\u7528\u591a\u500b\u6307\u5357<\/span>\nresults = mlflow.genai.evaluate(\n    data=eval_dataset,\n    predict_fn=my_app,\n    scorers=&#91;professional_tone, no_code_in_response],\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\">Guidelines Scorer \u6703\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u628a\u4f60\u7684\u6307\u5357\u8f49\u6210\u8a55\u4f30 prompt<\/li>\n<li>\u8b93 Judge \u6a21\u578b\u5224\u65b7 pass \u6216 fail<\/li>\n<li>\u63d0\u4f9b\u5224\u65b7\u7406\u7531<\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5efa\u7acb\u81ea\u8a02\u8a55\u4f30\u6307\u6a19\uff1amake_judge()<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u4f60\u9700\u8981\u66f4\u8907\u96dc\u7684\u8a55\u4f30\u908f\u8f2f\u6642\uff0c\u4f7f\u7528 <code>make_judge()<\/code> \u5efa\u7acb\u81ea\u8a02 Scorer\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=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.judges import make_judge\nfrom typing import Literal\n\n<span class=\"hljs-comment\"># \u5efa\u7acb\u4e00\u500b\u8a55\u4f30\u300c\u9023\u8cab\u6027\u300d\u7684 Judge<\/span>\ncoherence_judge = make_judge(\n    name=<span class=\"hljs-string\">\"coherence\"<\/span>,\n    instructions=(\n        <span class=\"hljs-string\">\"\u8a55\u4f30\u56de\u7b54\u662f\u5426\u9023\u8cab\uff0c\u4fdd\u6301\u4e00\u81f4\u7684\u8a9e\u8abf\uff0c\u4e26\u6709\u6e05\u695a\u7684\u908f\u8f2f\u6d41\u7a0b\u3002\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u554f\u984c: {{ inputs }}\\n\"<\/span>\n        <span class=\"hljs-string\">\"\u56de\u7b54: {{ outputs }}\\n\"<\/span>\n    ),\n    feedback_value_type=Literal&#91;<span class=\"hljs-string\">\"coherent\"<\/span>, <span class=\"hljs-string\">\"somewhat_coherent\"<\/span>, <span class=\"hljs-string\">\"incoherent\"<\/span>],\n    model=<span class=\"hljs-string\">\"openai:\/gpt-4o-mini\"<\/span>,\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\">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<h3 class=\"wp-block-heading\">Template \u8b8a\u6578<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><code>make_judge()<\/code> \u652f\u63f4\u4ee5\u4e0b\u4fdd\u7559\u8b8a\u6578\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u8b8a\u6578<\/th>\n<th>\u8aaa\u660e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><code>{{ inputs }}<\/code><\/td>\n<td>\u6a21\u578b\u7684\u8f38\u5165\u8cc7\u6599<\/td>\n<\/tr>\n<tr>\n<td><code>{{ outputs }}<\/code><\/td>\n<td>\u6a21\u578b\u7684\u8f38\u51fa\u7d50\u679c<\/td>\n<\/tr>\n<tr>\n<td><code>{{ expectations }}<\/code><\/td>\n<td>\u9810\u671f\u7b54\u6848\uff08ground truth\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>{{ trace }}<\/code><\/td>\n<td>\u5b8c\u6574\u57f7\u884c\u8ffd\u8e64\uff08\u4e0d\u80fd\u8207\u5176\u4ed6\u8b8a\u6578\u6df7\u7528\uff09<\/td>\n<\/tr>\n<tr>\n<td><code>{{ conversation }}<\/code><\/td>\n<td>\u591a\u8f2a\u5c0d\u8a71\u6b77\u53f2<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">feedback_value_type \u9078\u9805<\/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-7\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u5e03\u6797\u503c\uff1apass\/fail<\/span>\nmake_judge(..., feedback_value_type=bool)\n\n<span class=\"hljs-comment\"># \u6578\u503c\u8a55\u5206<\/span>\nmake_judge(..., feedback_value_type=int)    <span class=\"hljs-comment\"># \u6574\u6578<\/span>\nmake_judge(..., feedback_value_type=float)  <span class=\"hljs-comment\"># \u6d6e\u9ede\u6578<\/span>\n\n<span class=\"hljs-comment\"># \u5206\u985e\u7d50\u679c\uff08\u63a8\u85a6\uff09<\/span>\nmake_judge(..., feedback_value_type=Literal&#91;<span class=\"hljs-string\">\"good\"<\/span>, <span class=\"hljs-string\">\"acceptable\"<\/span>, <span class=\"hljs-string\">\"poor\"<\/span>])\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<h3 class=\"wp-block-heading\">\u5b8c\u6574\u7bc4\u4f8b\uff1a\u8a55\u4f30\u4e8b\u5be6\u6b63\u78ba\u6027<\/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-8\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.judges import make_judge\nfrom typing import Literal\n\nfactual_accuracy_judge = make_judge(\n    name=<span class=\"hljs-string\">\"factual_accuracy\"<\/span>,\n    instructions=<span class=\"hljs-string\">\"\"<\/span><span class=\"hljs-string\">\"\n    \u8a55\u4f30\u56de\u7b54\u7684\u4e8b\u5be6\u6b63\u78ba\u6027\u3002\n\n    \u53c3\u8003\u8cc7\u6599: {{ expectations }}\n\n    \u56de\u7b54: {{ outputs }}\n\n    \u6aa2\u67e5\u56de\u7b54\u4e2d\u7684\u4e8b\u5be6\u9673\u8ff0\u662f\u5426\u8207\u53c3\u8003\u8cc7\u6599\u4e00\u81f4\u3002\n    - \"<\/span>accurate<span class=\"hljs-string\">\": \u6240\u6709\u4e8b\u5be6\u9673\u8ff0\u90fd\u6b63\u78ba\n    - \"<\/span>partially_accurate<span class=\"hljs-string\">\": \u90e8\u5206\u6b63\u78ba\uff0c\u90e8\u5206\u932f\u8aa4\u6216\u907a\u6f0f\n    - \"<\/span>inaccurate<span class=\"hljs-string\">\": \u4e3b\u8981\u4e8b\u5be6\u932f\u8aa4\u6216\u5b58\u5728\u5e7b\u89ba\n    \"<\/span><span class=\"hljs-string\">\"\"<\/span>,\n    feedback_value_type=Literal&#91;<span class=\"hljs-string\">\"accurate\"<\/span>, <span class=\"hljs-string\">\"partially_accurate\"<\/span>, <span class=\"hljs-string\">\"inaccurate\"<\/span>],\n    model=<span class=\"hljs-string\">\"openai:\/gpt-4o-mini\"<\/span>,\n)\n\n<span class=\"hljs-comment\"># \u4f7f\u7528\u81ea\u8a02 Judge<\/span>\nresults = mlflow.genai.evaluate(\n    data=eval_dataset,\n    predict_fn=my_app,\n    scorers=&#91;factual_accuracy_judge],\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<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5f9e Production Traces \u5efa\u7acb\u8a55\u4f30\u8cc7\u6599\u96c6<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u6700\u5f37\u5927\u7684\u529f\u80fd\u4e4b\u4e00\u662f\u80fd\u5f9e\u5be6\u969b\u904b\u884c\u7684 Traces \u64f7\u53d6\u8a55\u4f30\u8cc7\u6599\u3002\u9019\u89e3\u6c7a\u4e86\u300c\u8a55\u4f30\u8cc7\u6599\u96c6\u4e0d\u5920\u771f\u5be6\u300d\u7684\u554f\u984c\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u6b65\u9a5f 1\uff1a\u641c\u5c0b Traces<\/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-9\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">import mlflow\nfrom datetime import datetime, timedelta\n\n<span class=\"hljs-comment\"># \u53d6\u5f97\u904e\u53bb 24 \u5c0f\u6642\u7684 traces<\/span>\nyesterday = datetime.now() - timedelta(days=<span class=\"hljs-number\">1<\/span>)\ntraces = mlflow.search_traces(\n    filter_string=f<span class=\"hljs-string\">\"timestamp &gt; {int(yesterday.timestamp() * 1000)}\"<\/span>\n)\n\n<span class=\"hljs-comment\"># traces \u662f pandas DataFrame<\/span>\n<span class=\"hljs-keyword\">print<\/span>(f<span class=\"hljs-string\">\"\u627e\u5230 {len(traces)} \u7b46 traces\"<\/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<h3 class=\"wp-block-heading\">\u6b65\u9a5f 2\uff1a\u76f4\u63a5\u8a55\u4f30 Traces<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u95dc\u9375\u4f86\u4e86\u2014\u2014\u4f60\u53ef\u4ee5\u76f4\u63a5\u628a traces \u4e1f\u9032 <code>evaluate()<\/code>\uff0c<strong>\u4e0d\u9700\u8981 predict_fn<\/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-10\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">from mlflow.genai.scorers import RetrievalGroundedness, RelevanceToQuery, Guidelines\n\n<span class=\"hljs-comment\"># \u5b9a\u7fa9 Scorers<\/span>\nscorers = &#91;\n    RetrievalGroundedness(),\n    RelevanceToQuery(),\n    Guidelines(\n        name=<span class=\"hljs-string\">\"professional_tone\"<\/span>,\n        guidelines=<span class=\"hljs-string\">\"\u56de\u7b54\u5fc5\u9808\u4fdd\u6301\u5c08\u696d\u8a9e\u8abf\"<\/span>\n    ),\n]\n\n<span class=\"hljs-comment\"># \u76f4\u63a5\u8a55\u4f30 traces<\/span>\nresults = mlflow.genai.evaluate(\n    data=traces,\n    scorers=scorers,\n)\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\">MLflow \u6703\u81ea\u52d5\u5f9e Trace \u7269\u4ef6\u4e2d\u63d0\u53d6 inputs\u3001outputs \u548c\u5176\u4ed6\u8cc7\u8a0a\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u70ba\u4ec0\u9ebc\u9019\u5f88\u91cd\u8981\uff1f<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u96e2\u7dda\u8a55\u4f30<\/strong>\uff1a\u4e0d\u7528\u91cd\u65b0\u8dd1\u6a21\u578b\uff0c\u7bc0\u7701\u6210\u672c<\/li>\n<li><strong>\u771f\u5be6\u8cc7\u6599<\/strong>\uff1a\u8a55\u4f30\u7684\u662f\u5be6\u969b\u7528\u6236\u4e92\u52d5<\/li>\n<li><strong>\u54c1\u8cea\u76e3\u63a7<\/strong>\uff1a\u5b9a\u671f\u8a55\u4f30 production traces \u4ee5\u767c\u73fe\u9000\u5316<\/li>\n<li><strong>\u5efa\u7acb Ground Truth<\/strong>\uff1a\u5f9e\u771f\u5be6\u6848\u4f8b\u4e2d\u6311\u9078\u597d\u7684\u56de\u7b54\u4f5c\u70ba\u6a19\u6e96<\/li>\n<\/ol>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u5be6\u6230\uff1a\u6bd4\u8f03\u4e0d\u540c Prompt \u7248\u672c<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u73fe\u5728\u4f86\u770b\u4e00\u500b\u5b8c\u6574\u7684\u5be6\u6230\u6848\u4f8b\uff1a\u6bd4\u8f03\u5169\u500b prompt \u7248\u672c\u7684\u6548\u679c\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u60c5\u5883\u8a2d\u5b9a<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5047\u8a2d\u4f60\u6709\u4e00\u500b QA \u61c9\u7528\uff0c\u60f3\u6e2c\u8a66\u5169\u7a2e\u4e0d\u540c\u7684 system prompt\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-11\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\">import mlflow\nfrom openai import OpenAI\n\nclient = OpenAI()\n\n<span class=\"hljs-comment\"># Prompt \u7248\u672c 1\uff1a\u7c21\u6f54\u98a8\u683c<\/span>\nPROMPT_V1 = &#91;\n    {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"system\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u4f60\u662f\u4e00\u500b\u52a9\u624b\u3002\u7c21\u6f54\u56de\u7b54\u554f\u984c\u3002\"<\/span>},\n    {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"user\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u554f\u984c: {question}\"<\/span>},\n]\n\n<span class=\"hljs-comment\"># Prompt \u7248\u672c 2\uff1a\u8a73\u7d30\u98a8\u683c<\/span>\nPROMPT_V2 = &#91;\n    {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"system\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u4f60\u662f\u4e00\u500b\u5c08\u696d\u7684\u6280\u8853\u52a9\u624b\u3002\u8acb\u8a73\u7d30\u89e3\u91cb\u6982\u5ff5\uff0c\u4e26\u63d0\u4f9b\u76f8\u95dc\u7bc4\u4f8b\u3002\"<\/span>},\n    {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"user\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u554f\u984c: {question}\"<\/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<h3 class=\"wp-block-heading\">\u5efa\u7acb\u8a55\u4f30\u8cc7\u6599\u96c6<\/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=\"JavaScript\" data-shcb-language-slug=\"javascript\"><span><code class=\"hljs language-javascript\">eval_dataset = &#91;\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u4ec0\u9ebc\u662f Docker\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {\n            <span class=\"hljs-string\">\"key_concepts\"<\/span>: &#91;<span class=\"hljs-string\">\"\u5bb9\u5668\"<\/span>, <span class=\"hljs-string\">\"\u865b\u64ec\u5316\"<\/span>, <span class=\"hljs-string\">\"\u6620\u50cf\u6a94\"<\/span>, <span class=\"hljs-string\">\"\u9694\u96e2\"<\/span>]\n        },\n    },\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"\u5982\u4f55\u5efa\u7acb REST API\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {\n            <span class=\"hljs-string\">\"key_concepts\"<\/span>: &#91;<span class=\"hljs-string\">\"HTTP\"<\/span>, <span class=\"hljs-string\">\"\u7aef\u9ede\"<\/span>, <span class=\"hljs-string\">\"\u8acb\u6c42\"<\/span>, <span class=\"hljs-string\">\"\u56de\u61c9\"<\/span>]\n        },\n    },\n    {\n        <span class=\"hljs-string\">\"inputs\"<\/span>: {<span class=\"hljs-string\">\"question\"<\/span>: <span class=\"hljs-string\">\"Git \u548c GitHub \u6709\u4ec0\u9ebc\u4e0d\u540c\uff1f\"<\/span>},\n        <span class=\"hljs-string\">\"expectations\"<\/span>: {\n            <span class=\"hljs-string\">\"key_concepts\"<\/span>: &#91;<span class=\"hljs-string\">\"\u7248\u672c\u63a7\u5236\"<\/span>, <span class=\"hljs-string\">\"\u672c\u5730\"<\/span>, <span class=\"hljs-string\">\"\u9060\u7aef\"<\/span>, <span class=\"hljs-string\">\"\u5e73\u53f0\"<\/span>]\n        },\n    },\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\">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<h3 class=\"wp-block-heading\">\u5b9a\u7fa9 Scorers<\/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\">from mlflow.genai.scorers import RelevanceToQuery, Completeness, Guidelines\nfrom mlflow.genai.judges import make_judge\nfrom typing import Literal\n\n<span class=\"hljs-comment\"># \u81ea\u8a02 Scorer\uff1a\u6982\u5ff5\u6db5\u84cb\u5ea6<\/span>\nconcept_coverage = make_judge(\n    name=<span class=\"hljs-string\">\"concept_coverage\"<\/span>,\n    instructions=<span class=\"hljs-string\">\"\"<\/span><span class=\"hljs-string\">\"\n    \u6aa2\u67e5\u56de\u7b54\u662f\u5426\u6db5\u84cb\u9810\u671f\u7684\u95dc\u9375\u6982\u5ff5\u3002\n\n    \u9810\u671f\u6982\u5ff5: {{ expectations }}\n    \u56de\u7b54: {{ outputs }}\n\n    \u8a55\u4f30\u56de\u7b54\u6db5\u84cb\u4e86\u591a\u5c11\u9810\u671f\u6982\u5ff5\u3002\n    \"<\/span><span class=\"hljs-string\">\"\"<\/span>,\n    feedback_value_type=Literal&#91;<span class=\"hljs-string\">\"full\"<\/span>, <span class=\"hljs-string\">\"partial\"<\/span>, <span class=\"hljs-string\">\"minimal\"<\/span>],\n    model=<span class=\"hljs-string\">\"openai:\/gpt-4o-mini\"<\/span>,\n)\n\nscorers = &#91;\n    RelevanceToQuery(),\n    Completeness(),\n    Guidelines(name=<span class=\"hljs-string\">\"is_professional\"<\/span>, guidelines=<span class=\"hljs-string\">\"\u56de\u7b54\u5fc5\u9808\u5c08\u696d\u4e14\u6709\u689d\u7406\"<\/span>),\n    concept_coverage,\n]\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<h3 class=\"wp-block-heading\">\u57f7\u884c\u6bd4\u8f03\u8a55\u4f30<\/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-14\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u7248\u672c 1 \u7684 predict function<\/span>\n@mlflow.trace\ndef predict_v1(question: str) -&gt; str:\n    messages = &#91;\n        {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"system\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u4f60\u662f\u4e00\u500b\u52a9\u624b\u3002\u7c21\u6f54\u56de\u7b54\u554f\u984c\u3002\"<\/span>},\n        {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"user\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: f<span class=\"hljs-string\">\"\u554f\u984c: {question}\"<\/span>},\n    ]\n    response = client.chat.completions.create(\n        model=<span class=\"hljs-string\">\"gpt-4o-mini\"<\/span>,\n        messages=messages,\n    )\n    <span class=\"hljs-keyword\">return<\/span> response.choices&#91;<span class=\"hljs-number\">0<\/span>].message.content\n\n<span class=\"hljs-comment\"># \u7248\u672c 2 \u7684 predict function<\/span>\n@mlflow.trace\ndef predict_v2(question: str) -&gt; str:\n    messages = &#91;\n        {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"system\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: <span class=\"hljs-string\">\"\u4f60\u662f\u4e00\u500b\u5c08\u696d\u7684\u6280\u8853\u52a9\u624b\u3002\u8acb\u8a73\u7d30\u89e3\u91cb\u6982\u5ff5\uff0c\u4e26\u63d0\u4f9b\u76f8\u95dc\u7bc4\u4f8b\u3002\"<\/span>},\n        {<span class=\"hljs-string\">\"role\"<\/span>: <span class=\"hljs-string\">\"user\"<\/span>, <span class=\"hljs-string\">\"content\"<\/span>: f<span class=\"hljs-string\">\"\u554f\u984c: {question}\"<\/span>},\n    ]\n    response = client.chat.completions.create(\n        model=<span class=\"hljs-string\">\"gpt-4o-mini\"<\/span>,\n        messages=messages,\n    )\n    <span class=\"hljs-keyword\">return<\/span> response.choices&#91;<span class=\"hljs-number\">0<\/span>].message.content\n\n<span class=\"hljs-comment\"># \u5206\u5225\u8a55\u4f30\u5169\u500b\u7248\u672c<\/span>\nwith mlflow.start_run(run_name=<span class=\"hljs-string\">\"prompt_v1_evaluation\"<\/span>):\n    results_v1 = mlflow.genai.evaluate(\n        data=eval_dataset,\n        predict_fn=predict_v1,\n        scorers=scorers,\n    )\n\nwith mlflow.start_run(run_name=<span class=\"hljs-string\">\"prompt_v2_evaluation\"<\/span>):\n    results_v2 = mlflow.genai.evaluate(\n        data=eval_dataset,\n        predict_fn=predict_v2,\n        scorers=scorers,\n    )\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-14\"><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<h3 class=\"wp-block-heading\">\u67e5\u770b\u7d50\u679c<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u8a55\u4f30\u5b8c\u6210\u5f8c\uff0c\u4f60\u53ef\u4ee5\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5728 MLflow UI \u67e5\u770b<\/strong>\uff1a\u6bd4\u8f03\u5169\u500b run \u7684\u6307\u6a19<\/li>\n<li><strong>\u7a0b\u5f0f\u5316\u5206\u6790<\/strong>\uff1a<\/li>\n<\/ol>\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-15\" data-shcb-language-name=\"PHP\" data-shcb-language-slug=\"php\"><span><code class=\"hljs language-php\"><span class=\"hljs-comment\"># \u53d6\u5f97\u8a55\u4f30\u7d50\u679c\u7684 DataFrame<\/span>\n<span class=\"hljs-keyword\">print<\/span>(<span class=\"hljs-string\">\"=== Prompt V1 \u7d50\u679c ===\"<\/span>)\n<span class=\"hljs-keyword\">print<\/span>(results_v1.tables&#91;<span class=\"hljs-string\">\"eval_results\"<\/span>])\n\n<span class=\"hljs-keyword\">print<\/span>(<span class=\"hljs-string\">\"\\n=== Prompt V2 \u7d50\u679c ===\"<\/span>)\n<span class=\"hljs-keyword\">print<\/span>(results_v2.tables&#91;<span class=\"hljs-string\">\"eval_results\"<\/span>])\n<\/code><\/span><small class=\"shcb-language\" id=\"shcb-language-15\"><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<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u8a55\u4f30\u7d50\u679c\u8996\u89ba\u5316<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow UI \u63d0\u4f9b\u591a\u7a2e\u8996\u89ba\u5316\u529f\u80fd\uff1a<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u8a55\u4f30\u8868\u683c<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u6bcf\u6b21\u8a55\u4f30\u6703\u7522\u751f\u4e00\u500b\u7d50\u679c\u8868\u683c\uff0c\u5305\u542b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6bcf\u7b46\u8cc7\u6599\u7684\u8f38\u5165\u3001\u8f38\u51fa<\/li>\n<li>\u6bcf\u500b Scorer \u7684\u5206\u6578\u548c\u7406\u7531<\/li>\n<li>\u5f59\u7e3d\u7d71\u8a08<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u6bd4\u8f03\u8996\u5716<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728 MLflow UI \u4e2d\uff0c\u4f60\u53ef\u4ee5\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u9078\u64c7\u591a\u500b\u8a55\u4f30 run<\/li>\n<li>\u4e26\u6392\u6bd4\u8f03\u6307\u6a19<\/li>\n<li>\u8ffd\u8e64\u7248\u672c\u9593\u7684\u6539\u5584\u6216\u9000\u5316<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">3. Trace \u8996\u5716<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u9ede\u64ca\u4efb\u4e00\u8a55\u4f30\u7d50\u679c\uff0c\u53ef\u4ee5\u6df1\u5165\u67e5\u770b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5b8c\u6574\u7684 Trace \u57f7\u884c\u6d41\u7a0b<\/li>\n<li>Judge \u7684\u8a55\u4f30\u7406\u7531<\/li>\n<li>\u8f38\u5165\u8f38\u51fa\u5c0d\u7167<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u6700\u4f73\u5be6\u8e10<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">1. \u9078\u64c7\u6b63\u78ba\u7684 Judge \u6a21\u578b<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u958b\u767c\u968e\u6bb5<\/strong>\uff1a\u4f7f\u7528\u5f37\u5927\u7684\u6a21\u578b\uff08GPT-4o\u3001Claude Opus\uff09\u767c\u73fe\u554f\u984c\u6a21\u5f0f<\/li>\n<li><strong>Production<\/strong>\uff1a\u5207\u63db\u5230\u8f03\u5c0f\u7684\u6a21\u578b\uff08GPT-4o-mini\u3001Claude Haiku\uff09\u964d\u4f4e\u6210\u672c<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">2. \u901a\u7528 vs \u5c08\u7528\u6307\u6a19<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\"><p>\u7814\u7a76\u986f\u793a\uff0c\u901a\u7528\u6307\u6a19\u5982\u300cHallucination\u300d\u6216\u300cToxicity\u300d\u5728\u5be6\u52d9\u4e0a\u6548\u679c\u6709\u9650\u3002\u6210\u529f\u7684\u505a\u6cd5\u662f\u5206\u6790\u771f\u5be6\u8cc7\u6599\u3001\u767c\u73fe domain-specific \u7684\u5931\u6557\u6a21\u5f0f\uff0c\u7136\u5f8c\u5f9e\u982d\u5b9a\u7fa9\u81ea\u8a02\u8a55\u4f30\u6a19\u6e96\u3002<\/p><\/blockquote>\n\n\n\n<h3 class=\"wp-block-heading\">3. \u4e8c\u5143\u8a55\u5206\u512a\u65bc\u6578\u503c\u8a55\u5206<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">MLflow \u5efa\u8b70\u4f7f\u7528\u4e8c\u5143\uff08pass\/fail\uff09\u6216\u5c11\u91cf\u985e\u5225\u7684\u8a55\u5206\uff0c\u800c\u975e 1-5 \u7684\u6578\u503c\u8a55\u5206\u3002\u539f\u56e0\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>LLM \u7522\u751f\u7684\u5224\u65b7\u66f4\u4e00\u81f4<\/li>\n<li>\u66f4\u5bb9\u6613\u8a2d\u5b9a\u95be\u503c<\/li>\n<li>\u6c7a\u7b56\u66f4\u660e\u78ba<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">4. \u8fed\u4ee3\u6539\u5584\u6d41\u7a0b<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u5f9e\u771f\u5be6\u7528\u6236\u4e92\u52d5\u4e2d\u6536\u96c6 Traces<\/li>\n<li>\u8b93 domain experts \u6a19\u8a3b\u54c1\u8cea\u554f\u984c<\/li>\n<li>\u5206\u6790\u5e38\u898b\u7684\u5931\u6557\u6a21\u5f0f<\/li>\n<li>\u5b9a\u7fa9\u91dd\u5c0d\u9019\u4e9b\u6a21\u5f0f\u7684 Scorers<\/li>\n<li>\u4f7f\u7528\u81ea\u52d5\u5c0d\u9f4a\u529f\u80fd\u6821\u6e96 Scorers<\/li>\n<\/ol>\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<p class=\"wp-block-paragraph\">\u672c\u7bc7\u4ecb\u7d39\u4e86 MLflow Evaluation \u7684\u5b8c\u6574\u529f\u80fd\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u6982\u5ff5<\/th>\n<th>\u8aaa\u660e<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>LLM-as-a-Judge<\/td>\n<td>\u7528 LLM \u8a55\u4f30 LLM \u8f38\u51fa\uff0c\u53ef\u898f\u6a21\u5316\u4e14\u4e00\u81f4<\/td>\n<\/tr>\n<tr>\n<td>Scorer<\/td>\n<td>\u5b9a\u7fa9\u8a55\u4f30\u6a19\u6e96\u7684\u5143\u4ef6<\/td>\n<\/tr>\n<tr>\n<td>\u5167\u5efa Scorer<\/td>\n<td>RelevanceToQuery\u3001Correctness\u3001Safety \u7b49<\/td>\n<\/tr>\n<tr>\n<td>Guidelines<\/td>\n<td>\u7528\u81ea\u7136\u8a9e\u8a00\u5b9a\u7fa9 pass\/fail \u689d\u4ef6<\/td>\n<\/tr>\n<tr>\n<td>make_judge()<\/td>\n<td>\u5efa\u7acb\u81ea\u8a02\u8a55\u4f30\u6307\u6a19<\/td>\n<\/tr>\n<tr>\n<td>Trace \u8a55\u4f30<\/td>\n<td>\u5f9e production \u64f7\u53d6\u771f\u5be6\u8cc7\u6599\u8a55\u4f30<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u900f\u904e\u9019\u4e9b\u5de5\u5177\uff0c\u4f60\u53ef\u4ee5\u5efa\u7acb\u4e00\u500b\u5b8c\u6574\u7684\u54c1\u8cea\u4fdd\u8b49\u6d41\u7a0b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u958b\u767c\u6642\u5feb\u901f\u8fed\u4ee3 prompt<\/li>\n<li>\u4e0a\u7dda\u524d\u9a57\u8b49\u54c1\u8cea<\/li>\n<li>Production \u74b0\u5883\u6301\u7e8c\u76e3\u63a7<\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u7cfb\u5217\u7e3d\u7d50<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u606d\u559c\u4f60\u5b8c\u6210\u4e86\u3010MLflow GenAI \u6559\u5b78\u3011\u7cfb\u5217\uff01\u8b93\u6211\u5011\u56de\u9867\u6574\u500b\u7cfb\u5217\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table>\n<thead>\n<tr>\n<th>\u7bc7\u7ae0<\/th>\n<th>\u4e3b\u984c<\/th>\n<th>\u6838\u5fc3\u80fd\u529b<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>#01<\/td>\n<td>\u5feb\u901f\u958b\u59cb<\/td>\n<td>\u5b89\u88dd\u8a2d\u5b9a\u3001\u57fa\u672c\u6982\u5ff5<\/td>\n<\/tr>\n<tr>\n<td>#02<\/td>\n<td>Tracing<\/td>\n<td>\u8ffd\u8e64 LLM \u57f7\u884c\u6d41\u7a0b\u3001Debug<\/td>\n<\/tr>\n<tr>\n<td>#03<\/td>\n<td>Prompt Engineering<\/td>\n<td>\u7248\u672c\u63a7\u5236\u3001Prompt Registry<\/td>\n<\/tr>\n<tr>\n<td>#04<\/td>\n<td>Evaluation<\/td>\n<td>\u81ea\u52d5\u5316\u54c1\u8cea\u8a55\u4f30<\/td>\n<\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u9019\u56db\u500b\u529f\u80fd\u5f62\u6210\u4e00\u500b\u5b8c\u6574\u7684 GenAI \u958b\u767c\u5faa\u74b0\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\">\u8a2d\u8a08 Prompt \u2192 \u8ffd\u8e64\u57f7\u884c \u2192 \u8a55\u4f30\u54c1\u8cea \u2192 \u6539\u5584 Prompt \u2192 ...\n    \u2191                                    |\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n<\/code><\/span><\/pre><\/div>\n\n\n<p class=\"wp-block-paragraph\">MLflow GenAI \u8b93\u9019\u500b\u5faa\u74b0\u53ef\u4ee5\u88ab\u7cfb\u7d71\u5316\u3001\u53ef\u8ffd\u8e64\u3001\u53ef\u91cf\u5316\u3002\u795d\u4f60\u958b\u767c\u9806\u5229\uff01<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h2 class=\"wp-block-heading\">\u53c3\u8003\u8cc7\u6e90<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/\">MLflow Evaluating LLMs\/Agents<\/a><\/li>\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/scorers\/llm-judge\/predefined\">Predefined LLM Judge Scorers<\/a><\/li>\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/scorers\/llm-judge\/\">LLM-based Scorers (LLM-as-a-Judge)<\/a><\/li>\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/scorers\/llm-judge\/make-judge\/\">Using make_judge for Custom LLM Evaluation<\/a><\/li>\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/running-evaluation\/traces\/\">Evaluating Production Traces<\/a><\/li>\n<li><a href=\"https:\/\/mlflow.org\/docs\/latest\/genai\/eval-monitor\/running-evaluation\/prompts\/\">Evaluating Prompts<\/a><\/li>\n<\/ul>\n\n\n\n<div id=\"quiz-41c059b2\">\n<style>\n#quiz-41c059b2 .badge {\n        display: inline-block;\n        padding: 4px 10px;\n        border-radius: 12px;\n        font-size: 0.75em;\n        font-weight: bold;\n        margin-left: 8px;\n      }#quiz-41c059b2 .badge.scenario {\n        background: #ebf8ff;\n        color: #2b6cb0;\n      }#quiz-41c059b2 .badge.debug {\n        background: #fef3c7;\n        color: #92400e;\n      }#quiz-41c059b2 .question-adv {\n        background: white;\n        border-radius: 10px;\n        padding: 24px;\n        margin: 20px 0;\n        box-shadow: 0 2px 8px rgba(0, 0, 0, 0.08);\n        border-left: 4px solid #ed8936;\n      }#quiz-41c059b2 .question-adv h3 {\n        margin-top: 0;\n        color: #2d3748;\n      }#quiz-41c059b2 .code-block {\n        background: #1e1e1e;\n        color: #d4d4d4;\n        padding: 14px;\n        border-radius: 6px;\n        font-family: \"Menlo\", \"Monaco\", \"Courier New\", monospace;\n        font-size: 0.9em;\n        overflow-x: auto;\n        margin: 12px 0;\n        white-space: pre-wrap;\n      }#quiz-41c059b2 .options-adv {\n        list-style: none;\n        padding: 0;\n      }#quiz-41c059b2 .options-adv li {\n        padding: 14px 16px;\n        margin: 10px 0;\n        border: 2px solid #e2e8f0;\n        border-radius: 8px;\n        cursor: pointer;\n        transition: all 0.2s;\n      }#quiz-41c059b2 .options-adv li:hover:not(.disabled) {\n        border-color: #ed8936;\n        background: #fffaf0;\n      }#quiz-41c059b2 .options-adv li.disabled {\n        cursor: default;\n        opacity: 0.7;\n      }#quiz-41c059b2 .options-adv li.show-answer {\n        border-color: #48bb78;\n        background: #f0fff4;\n      }#quiz-41c059b2 .options-adv li.wrong {\n        border-color: #fc8181;\n        background: #fff5f5;\n      }#quiz-41c059b2 .feedback-adv {\n        margin-top: 16px;\n        padding: 14px;\n        border-radius: 8px;\n        display: none;\n        line-height: 1.6;\n      }#quiz-41c059b2 .feedback-adv.correct {\n        background: #f0fff4;\n        color: #276749;\n        display: block;\n      }#quiz-41c059b2 .feedback-adv.wrong {\n        background: #fff5f5;\n        color: #c53030;\n        display: block;\n      }#quiz-41c059b2 .feedback-adv \n      .score-adv {\n        text-align: center;\n        font-size: 1.4em;\n        padding: 24px;\n        background: white;\n        border-radius: 10px;\n        margin-top: 24px;\n        box-shadow: 0 2px 8px rgba(0, 0, 0, 0.08);\n      }#quiz-41c059b2 .intro {\n        background: #ebf8ff;\n        padding: 16px;\n        border-radius: 8px;\n        margin-bottom: 20px;\n        color: #2c5282;\n      }\n<\/style>\n<h1>\u9032\u968e\u6e2c\u9a57\uff1aMLflow GenAI \u6559\u5b78 &#8211; Evaluation<\/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 MLflow Evaluation \u529f\u80fd\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\u6b63\u5728\u958b\u767c\u4e00\u500b\u5ba2\u670d\u804a\u5929\u6a5f\u5668\u4eba\uff0c\u9700\u8981\u78ba\u4fdd\u56de\u7b54\u4e0d\u6703\u5305\u542b\u6709\u5bb3\u5167\u5bb9\uff0c\u4e14\u56de\u7b54\u8981\u5207\u984c\u3002\u4f60\u4e0d\u9700\u8981\u9810\u5148\u6e96\u5099\u6a19\u6e96\u7b54\u6848\uff0c\u61c9\u8a72\u9078\u7528\u54ea\u4e9b Scorer\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(1, 'A', 'C')\">A. <code>Correctness<\/code> \u548c <code>Equivalence<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(1, 'B', 'C')\">B. <code>Guidelines<\/code> \u548c <code>Completeness<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(1, 'C', 'C')\">C. <code>Safety<\/code> \u548c <code>RelevanceToQuery<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(1, 'D', 'C')\">D. <code>RetrievalGroundedness<\/code> \u548c <code>RetrievalRelevance<\/code><\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv1\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv2\">\n      <h3>2. \u540c\u4e8b\u5beb\u4e86\u4ee5\u4e0b\u8a55\u4f30\u7a0b\u5f0f\u78bc\uff0c\u4f46\u57f7\u884c\u5f8c <code>my_app<\/code> \u6c92\u6709\u88ab\u547c\u53eb\u3002\u554f\u984c\u51fa\u5728\u54ea\u88e1\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">eval_dataset = [\n    {&#8220;question&#8221;: &#8220;MLflow \u53ef\u4ee5\u7ba1\u7406 prompt \u55ce\uff1f&#8221;},\n    {&#8220;question&#8221;: &#8220;\u4ec0\u9ebc\u662f Tracing\uff1f&#8221;},\n]\n\ndef my_app(question: str) -> str:\n    return &#8220;\u9019\u662f\u56de\u7b54&#8221;\n\nresults = mlflow.genai.evaluate(\n    data=eval_dataset,\n    predict_fn=my_app,\n    scorers=[RelevanceToQuery()],\n)<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(2, 'A', 'B')\">A. \u7f3a\u5c11 <code>import mlflow<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(2, 'B', 'B')\">B. \u8cc7\u6599\u96c6\u683c\u5f0f\u932f\u8aa4\uff0c\u61c9\u8a72\u7528 <code>{\"inputs\": {\"question\": \"...\"}}<\/code> \u7d50\u69cb<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'C', 'B')\">C. <code>my_app<\/code> \u51fd\u5f0f\u61c9\u8a72\u662f async \u51fd\u5f0f<\/li>\n        <li onclick=\"checkAnswerAdv(2, 'D', 'B')\">D. \u7f3a\u5c11 <code>mlflow.start_run()<\/code><\/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\u7684 RAG \u61c9\u7528\u6709\u6642\u6703\u7522\u751f\u5e7b\u89ba\uff08\u56de\u7b54\u4e2d\u5305\u542b\u6aa2\u7d22\u8cc7\u6599\u6c92\u6709\u7684\u5167\u5bb9\uff09\u3002\u4f60\u60f3\u5efa\u7acb\u81ea\u52d5\u5316\u6aa2\u6e2c\u6a5f\u5236\uff0c\u61c9\u8a72\u4f7f\u7528\u54ea\u500b Scorer\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(3, 'A', 'D')\">A. <code>Correctness<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(3, 'B', 'D')\">B. <code>Safety<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(3, 'C', 'D')\">C. <code>RelevanceToQuery<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(3, 'D', 'D')\">D. <code>RetrievalGroundedness<\/code><\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv3\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv4\">\n      <h3>4. \u4f60\u60f3\u5efa\u7acb\u4e00\u500b\u81ea\u8a02\u8a55\u4f30\u6307\u6a19\uff0c\u7528\u4e09\u500b\u7b49\u7d1a\u8a55\u4f30\u56de\u7b54\u7684\u9023\u8cab\u6027\u3002\u4e0b\u9762\u7684 <code>make_judge()<\/code> \u7a0b\u5f0f\u78bc\u6709\u4ec0\u9ebc\u554f\u984c\uff1f <span class=\"badge debug\">\u932f\u8aa4\u8a3a\u65b7<\/span><\/h3>\n      <div class=\"code-block\">from mlflow.genai.judges import make_judge\n\ncoherence_judge = make_judge(\n    name=&#8221;coherence&#8221;,\n    instructions=&#8221;\u8a55\u4f30\u56de\u7b54\u662f\u5426\u9023\u8cab\u3002&#8221;,\n    feedback_value_type=[&#8220;coherent&#8221;, &#8220;somewhat_coherent&#8221;, &#8220;incoherent&#8221;],\n    model=&#8221;openai:\/gpt-4o-mini&#8221;,\n)<\/div>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(4, 'A', 'C')\">A. <code>name<\/code> \u53c3\u6578\u4e0d\u80fd\u6709\u5e95\u7dda<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'B', 'C')\">B. <code>instructions<\/code> \u6c92\u6709\u4f7f\u7528 <code>{{ outputs }}<\/code> \u8b8a\u6578<\/li>\n        <li onclick=\"checkAnswerAdv(4, 'C', 'C')\">C. <code>feedback_value_type<\/code> \u61c9\u8a72\u7528 <code>Literal[\"coherent\", \"somewhat_coherent\", \"incoherent\"]<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(4, 'D', 'C')\">D. <code>model<\/code> \u53c3\u6578\u683c\u5f0f\u932f\u8aa4<\/li>\n      <\/ul>\n      <div class=\"feedback-adv\" id=\"fbAdv4\"><\/div>\n    <\/div>\n\n    <div class=\"question-adv\" id=\"qAdv5\">\n      <h3>5. \u4f60\u60f3\u5f9e Production \u74b0\u5883\u6536\u96c6\u7684 Traces \u4e2d\u8a55\u4f30\u54c1\u8cea\uff0c\u800c\u4e0d\u60f3\u91cd\u65b0\u57f7\u884c\u6a21\u578b\u3002\u4e0b\u9762\u54ea\u7a2e\u505a\u6cd5\u6b63\u78ba\uff1f <span class=\"badge scenario\">\u60c5\u5883\u984c<\/span><\/h3>\n      <ul class=\"options-adv\">\n        <li onclick=\"checkAnswerAdv(5, 'A', 'B')\">A. \u5fc5\u9808\u63d0\u4f9b <code>predict_fn<\/code>\uff0c\u7121\u6cd5\u76f4\u63a5\u8a55\u4f30 Traces<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'B', 'B')\">B. \u4f7f\u7528 <code>mlflow.search_traces()<\/code> \u53d6\u5f97 Traces \u5f8c\uff0c\u76f4\u63a5\u50b3\u5165 <code>mlflow.genai.evaluate(data=traces, scorers=...)<\/code><\/li>\n        <li onclick=\"checkAnswerAdv(5, 'C', 'B')\">C. \u9700\u8981\u5148\u5c07 Traces \u8f49\u63db\u6210 CSV \u683c\u5f0f\u518d\u8a55\u4f30<\/li>\n        <li onclick=\"checkAnswerAdv(5, 'D', 'B')\">D. \u53ea\u6709 CLI \u6a21\u5f0f\u624d\u80fd\u8a55\u4f30 Production Traces<\/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: \"<code>Safety<\/code> \u7528\u65bc\u6aa2\u6e2c\u6709\u5bb3\u5167\u5bb9\uff0c<code>RelevanceToQuery<\/code> \u7528\u65bc\u8a55\u4f30\u56de\u7b54\u662f\u5426\u5207\u984c\u3002\u9019\u5169\u500b Scorer \u90fd\u4e0d\u9700\u8981 Ground Truth\u3002<code>Correctness<\/code> \u548c <code>Equivalence<\/code> \u9700\u8981\u9810\u671f\u7b54\u6848\uff0c<code>RetrievalGroundedness<\/code> \u662f\u91dd\u5c0d RAG \u61c9\u7528\u7684\u6aa2\u7d22\u4f9d\u64da\u8a55\u4f30\u3002\",\n        2: \"\u8cc7\u6599\u96c6\u4e2d\u6bcf\u7b46\u8cc7\u6599\u61c9\u8a72\u4f7f\u7528 <code>{\\\"inputs\\\": {\\\"question\\\": \\\"...\\\"}}<\/code> \u7d50\u69cb\uff0c\u800c\u4e0d\u662f\u76f4\u63a5\u7528 <code>{\\\"question\\\": \\\"...\\\"}<\/code>\u3002<code>inputs<\/code> \u662f dict \u683c\u5f0f\uff0c\u6703\u88ab\u50b3\u7d66 predict function\u3002\",\n        3: \"<code>RetrievalGroundedness<\/code> \u5c08\u9580\u7528\u65bc\u8a55\u4f30\u300c\u56de\u7b54\u662f\u5426\u6709\u6aa2\u7d22\u4f9d\u64da\u300d\uff0c\u53ef\u4ee5\u5075\u6e2c\u56de\u7b54\u4e2d\u662f\u5426\u5305\u542b\u6aa2\u7d22\u8cc7\u6599\u6c92\u6709\u7684\u5167\u5bb9\uff08\u5e7b\u89ba\uff09\u3002\u9019\u662f RAG \u61c9\u7528\u54c1\u8cea\u8a55\u4f30\u7684\u91cd\u8981\u6307\u6a19\u3002\",\n        4: \"<code>feedback_value_type<\/code> \u9700\u8981\u4f7f\u7528 Python \u7684 <code>Literal<\/code> \u985e\u578b\u4f86\u5b9a\u7fa9\u5206\u985e\u7d50\u679c\uff0c\u4f8b\u5982 <code>Literal[\\\"coherent\\\", \\\"somewhat_coherent\\\", \\\"incoherent\\\"]<\/code>\u3002\u76f4\u63a5\u50b3\u5165 list \u662f\u932f\u8aa4\u7684\u8a9e\u6cd5\u3002\",\n        5: \"MLflow \u652f\u63f4\u76f4\u63a5\u8a55\u4f30 Traces\uff0c\u4e0d\u9700\u8981 <code>predict_fn<\/code>\u3002\u4f7f\u7528 <code>mlflow.search_traces()<\/code> \u53d6\u5f97 Traces \u5f8c\uff0c\u53ef\u4ee5\u76f4\u63a5\u50b3\u5165 <code>evaluate()<\/code>\u3002MLflow \u6703\u81ea\u52d5\u5f9e Trace \u7269\u4ef6\u4e2d\u63d0\u53d6 inputs\u3001outputs \u7b49\u8cc7\u8a0a\uff0c\u5be6\u73fe\u96e2\u7dda\u8a55\u4f30\u3002\"\n      };\n\n      function checkAnswerAdv(qNum, selected, correctAns) {\n        const question = document.getElementById(\"qAdv\" + qNum);\n        if (question.classList.contains(\"answered\")) return;\n\n        question.classList.add(\"answered\");\n        answeredAdv++;\n\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\n        const feedback = document.getElementById(\"fbAdv\" + qNum);\n        if (selected === correctAns) {\n          correctAdv++;\n          feedback.className = \"feedback-adv correct\";\n          feedback.innerHTML = \"&#10003; \u6b63\u78ba\uff01\" + explanationsAdv[qNum];\n        } else {\n          feedback.className = \"feedback-adv wrong\";\n          feedback.innerHTML =\n            \"&#10007; \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\n        if (answeredAdv === totalAdv) {\n          const scoreDiv = document.getElementById(\"scoreAdv\");\n          scoreDiv.style.display = \"block\";\n          let msg = \"\";\n          if (correctAdv === totalAdv) msg = \"\u5b8c\u7f8e\uff01\u4f60\u5df2\u7d93\u6e96\u5099\u597d\u5be6\u6230\u4e86\uff01\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.8))\n            msg = \"\u5f88\u68d2\uff01\u518d\u8907\u7fd2\u4e00\u4e0b\u932f\u7684\u984c\u76ee\";\n          else if (correctAdv >= Math.ceil(totalAdv * 0.6))\n            msg = \"\u5efa\u8b70\u91cd\u8b80\u76f8\u95dc\u6bb5\u843d\u5f8c\u518d\u8a66\u4e00\u6b21\";\n          else msg = \"\u5efa\u8b70\u91cd\u65b0\u95b1\u8b80\u6587\u4ef6\uff0c\u7279\u5225\u6ce8\u610f Scorer \u7684\u9078\u7528\u5834\u666f\";\n          scoreDiv.innerHTML =\n            \"\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\uff1aMLflow GenAI \u6559\u5b78 &#8211; Evaluation \u5171 5 \u984c\uff0c\u9ede\u9078\u7b54\u6848\u5f8c\u6703\u7acb\u5373\u986f\u793a\u7d50 [&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":[215,206,223,222],"tags":[401,393,402,392,403],"class_list":["post-443","post","type-post","status-publish","format-standard","hentry","category-ai-ml","category-l3-","category-mlflow","category-mlops","tag-evaluation","tag-genai","tag-llm-as-judge","tag-mlflow","tag-403"],"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\/443","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=443"}],"version-history":[{"count":0,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/posts\/443\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/media?parent=443"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/categories?post=443"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.itsmygo.uk\/wordpress\/wp-json\/wp\/v2\/tags?post=443"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}