{"id":91840,"date":"2026-08-08T15:00:52","date_gmt":"2026-08-08T07:00:52","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/91840.html"},"modified":"2026-08-08T15:00:52","modified_gmt":"2026-08-08T07:00:52","slug":"%e5%9f%ba%e4%ba%8eopenevals%e7%9a%84%e8%87%aa%e5%8a%a8%e5%8c%96%e8%af%84%e4%bc%b0-06%e8%af%84%e4%bc%b0agent%e8%be%93%e5%87%ba%e6%96%87%e6%9c%ac%e7%9a%84%e8%b4%a8%e9%87%8f%e4%b8%8a%e7%af%87","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/91840.html","title":{"rendered":"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7]"},"content":{"rendered":"

\u5bf9\u4e8e\u6211\u4eec\u4e2a\u4eba\u6765\u8bf4，Agent\u7f16\u7a0b\u6700\u5927\u7684\u6311\u6218\u4e0d\u662f\u5404\u79cd\u6240\u8c13\u7684\u5de5\u4f5c，\u800c\u662f\u5199\u63d0\u793a\u8bcd，\u6bd5\u7adf\u6211\u4e0d\u662f\u4e13\u4e1a\u7684\u6587\u5b57\u5de5\u4f5c\u8005\u3002OpenEvals\u9488\u5bf9\u4e00\u7cfb\u5217\u9488\u5bf9\u8f93\u51fa\u6587\u672c\u8d28\u91cf\u7684\u8bc4\u4f30\u6307\u6807\u5b9a\u4e49\u4e86\u5bf9\u5e94\u7684\u63d0\u793a\u8bcd，\u7ed3\u5408\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-02：LLM-as-a-Judge:\u8ba9LLM\u5f53\u88c1\u5224\u6765\u8bc4\u4f30Agent\u7684\u8f93\u51fa\u4ecb\u7ecd\u4e86\u57fa\u4e8eLLM-as-a-Judge\u7684\u8bc4\u4f30\u5668，\u6211\u4eec\u53ef\u4ee5\u8fdb\u884c\u9488\u5bf9\u7b80\u6d01\u5ea6\u3001\u6b63\u786e\u6027\u3001\u5e7b\u89c9\u5ea6\u3001\u76f8\u5173\u6027\u3001\u8ba1\u5212\u9075\u5faa\u5ea6\u3001\u4ee3\u7801\u6b63\u786e\u6027\u7b49\u6307\u6807\u7684\u8bc4\u4f30\u3002\u6211\u4eec\u4f1a\u5206\u4e0a\u4e0b\u6587\u4e24\u7bc7\u6587\u7ae0\u6765\u4ecb\u7ecd\u9488\u5bf9\u8fd9\u4e9b\u9884\u5b9a\u4e49\u63d0\u793a\u8bcd\u7684\u4f7f\u7528\u3002<\/p>\n

1. \u7b80\u6d01\u5ea6（CONCISENES）<\/h3>\n

\u7b80\u6d01\u5ea6\u8bc4\u4f30\u65e8\u5728\u786e\u5b9a\u8f93\u51fa\u662f\u5426\u8db3\u591f\u7b80\u7ec3，\u662f\u5426\u907f\u514d\u4e86\u4e0d\u5fc5\u8981\u7684\u5e9f\u8bdd\u548c\u586b\u5145\u5185\u5bb9，\u7528\u6237\u68c0\u67e5Agent\u662f\u5426\u6b63\u9762\u56de\u7b54，\u6709\u6ca1\u6709\u7ed9\u51fa\u4e00\u5927\u5806\u6ca1\u6709\u4ef7\u503c\u7684\u5ba2\u5957\u8bdd\u6216\u91cd\u590d\u957f\u53e5\u3002OpenEvals\u9884\u5b9a\u7684\u9488\u5bf9\u8f93\u51fa\u8d28\u91cf\u7684\u63d0\u793a\u8bcd\u90fd\u5b9a\u4e49\u5728openevals.propmpts.quality\u8def\u5f84\u4e0b，\u7b80\u6d01\u6027\u8bc4\u4f30\u63d0\u793a\u8bcd\u5bf9\u5e94\u7684\u6210\u5458\u540d\u79f0\u4e3aANSWER_RELEVANCE_PROMPT，\u5177\u4f53\u5185\u5bb9\u5982\u4e0b\u3002\u4ece\u63d0\u793a\u8bcd\u6a21\u677f\u7684\u5b9a\u4e49\u53ef\u4ee5\u770b\u51fa，\u7531\u4e8e\u5b9a\u4e49\u4e86{inputs}\u548c{outputs}\u5360\u4f4d\u7b26，\u6240\u4ee5\u5728\u6267\u884c\u7684\u5bf9\u5e94\u8bc4\u4f30\u5668\u7684\u65f6\u5019\u9700\u8981\u6307\u5b9a\u8f93\u5165\u3002<\/p>\n

You are an expert evaluator assessing whether outputs are relevant to the given input. Your task is to determine whether EACH statement appropriately addresses what was asked.<\/p>\n

<Rubric>
\nA relevant output:
\n– Directly answers the question or addresses the request
\n– Provides information specifically asked for
\n– Stays on topic with the input's intent
\n– Contributes meaningfully to fulfilling the request<\/p>\n

An irrelevant output:
\n– Discusses topics not requested or implied by the input
\n– Provides unnecessary tangents or digressions
\n– Includes information that doesn't answer the question
\n– Addresses a different question than what was asked
\n<\/Rubric><\/p>\n

<Instructions>
\nFor each output:
\n– Read the original input carefully to understand what was asked
\n– Examine the output and identify its core claim or purpose
\n– Determine if the output directly addresses the input's request
\n– Assess whether the information helps fulfill what was asked
\n– Determine the answer relevancy of output and output a score
\n<\/Instructions><\/p>\n

<Reminder>
\nFocus on whether each statement helps answer the specific input question, not whether the statement is true or well-written. A statement can be factually correct but still irrelevant if it doesn't address what was asked.
\n<\/Reminder><\/p>\n

Now, grade the following example according to the above instructions:<\/p>\n

<example>
\n<input>
\n{inputs}
\n<\/input><\/p>\n

<output>
\n{outputs}
\n<\/output>
\n<\/example><\/p>\n

\u4ee5\u5982\u4e0b\u8fd9\u4e2a\u9488\u5bf9\u8f93\u51fa\u7b80\u6d01\u5ea6\u7684\u8bc4\u4f30\u4e3a\u4f8b，\u6211\u4eec\u9488\u5bf9\u540c\u4e00\u4e2a\u95ee\u9898：\u4ec0\u4e48\u662f\u5149\u5408\u4f5c\u7528？，\u63d0\u4f9b\u4e86\u4e24\u7aef\u7528\u4e8e\u8bc4\u4f30\u7684\u8f93\u51fa\u6587\u672c，\u524d\u8005\u5197\u957f\u3001\u5305\u542b\u5927\u91cf\u5e9f\u8bdd\/\u8f66\u8f71\u8f98\u8bdd，\u540e\u8005\u7b80\u6d01\u3001\u76f4\u51fb\u6838\u5fc3\u3001\u65e0\u591a\u4f59\u586b\u5145\u3002\u6240\u4ee5\u524d\u8005\u5f97\u4e860.15\u5206，\u540e\u8005\u5f97\u4e860.95\u98ce\u3002<\/p>\n

from<\/span> openevals import<\/span> create_async_llm_as_judge
\nfrom<\/span> langchain_openai import<\/span> ChatOpenAI
\nfrom<\/span> dotenv import<\/span> load_dotenv
\nfrom<\/span> openevals.<\/span>prompts.<\/span>quality import<\/span> CONCISENESS_PROMPT
\nimport<\/span> asyncio,<\/span>json,<\/span> functools<\/p>\n

load_dotenv(<\/span>)<\/span><\/p>\n

judge =<\/span> ChatOpenAI(<\/span>model=<\/span>"gpt-5.4-mini"<\/span>)<\/span>
\nasync<\/span> def<\/span> eval<\/span>(<\/span> outputs:<\/span>str<\/span>,<\/span> *<\/span>,<\/span> prompt:<\/span> str<\/span>,<\/span>inputs:<\/span>str<\/span>|<\/span>None<\/span> =<\/span> None<\/span>,<\/span> reference_outputs:<\/span>str<\/span>|<\/span>None<\/span> =<\/span> None<\/span>,<\/span>,<\/span>**<\/span>kwargs )<\/span>:<\/span>
\n evaluator =<\/span> create_async_llm_as_judge(<\/span>
\n prompt =<\/span> prompt,<\/span>
\n continuous=<\/span> True<\/span>,<\/span>
\n judge=<\/span> judge)<\/span>
\n result =<\/span> await<\/span> evaluator(<\/span>
\n inputs=<\/span> inputs,<\/span>
\n outputs=<\/span>outputs,<\/span>
\n reference_outputs=<\/span>reference_outputs,<\/span>
\n **<\/span>kwargs
\n )<\/span>
\n print<\/span>(<\/span>json.<\/span>dumps(<\/span>result,<\/span>ensure_ascii=<\/span>False<\/span>,<\/span> indent=<\/span>2<\/span>)<\/span>)<\/span><\/p>\n

async<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
\n inputs =<\/span>"\u4ec0\u4e48\u662f\u5149\u5408\u4f5c\u7528？"<\/span>
\n outputs1 =<\/span> """\\\\
\n\u8fd9\u662f\u4e00\u4e2a\u975e\u5e38\u597d\u7684\u95ee\u9898！
\n\u5728\u6211\u4eec\u7684\u81ea\u7136\u754c\u4e2d，\u690d\u7269\u626e\u6f14\u7740\u975e\u5e38\u91cd\u8981\u7684\u89d2\u8272，\u5b83\u4eec\u9700\u8981\u751f\u957f\u548c\u751f\u5b58\u3002
\n\u5173\u4e8e\u60a8\u95ee\u7684\u4ec0\u4e48\u662f\u5149\u5408\u4f5c\u7528，\u7b80\u5355\u6765\u8bf4，\u5149\u5408\u4f5c\u7528\u5176\u5b9e\u5c31\u662f\u690d\u7269\u3001\u85fb\u7c7b\u4ee5\u53ca\u67d0\u4e9b\u7279\u5b9a\u7684\u7ec6\u83cc，\u5728\u5b83\u4eec\u4f53\u5185\u7684\u53f6\u7eff\u4f53\u4e2d，\u901a\u8fc7\u5229\u7528\u592a\u9633\u7684\u5149\u80fd，\u6765\u628a\u4ece\u7a7a\u6c14\u4e2d\u5438\u6536\u7684\u6c34\u5206\u548c\u4e8c\u6c27\u5316\u78b3，\u901a\u8fc7\u4e00\u7cfb\u5217\u590d\u6742\u7684\u3001\u795e\u5947\u7684\u751f\u7269\u5316\u5b66\u53cd\u5e94\u8fc7\u7a0b，\u8f6c\u5316\u4e3a\u5b83\u4eec\u751f\u957f\u6240\u9700\u7684\u80fd\u91cf\u548c\u6709\u673a\u7269（\u6bd4\u5982\u8461\u8404\u7cd6），\u5e76\u4e14\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u5f53\u4e2d，\u5b83\u4eec\u8fd8\u4f1a\u628a\u6c27\u6c14\u91ca\u653e\u5230\u6211\u4eec\u547c\u5438\u7684\u7a7a\u6c14\u4e2d\u3002
\n\u603b\u800c\u8a00\u4e4b，\u8fd9\u5c31\u662f\u4e00\u4e2a\u80fd\u91cf\u8f6c\u6362\u7684\u5947\u5999\u8fc7\u7a0b，\u5bf9\u5730\u7403\u751f\u6001\u975e\u5e38\u5173\u952e\u3002
\n\u5e0c\u671b\u8fd9\u4e2a\u8be6\u7ec6\u7684\u89e3\u91ca\u5bf9\u60a8\u6709\u6240\u542f\u53d1\u548c\u5e2e\u52a9！
\n"""<\/span>
\n outputs2 =<\/span> "\u5149\u5408\u4f5c\u7528\u662f\u690d\u7269\u3001\u85fb\u7c7b\u548c\u67d0\u4e9b\u7ec6\u83cc\u5229\u7528\u5149\u80fd，\u5c06\u4e8c\u6c27\u5316\u78b3\u548c\u6c34\u8f6c\u5316\u4e3a\u6709\u673a\u7269（\u5982\u8461\u8404\u7cd6）\u5e76\u91ca\u653e\u6c27\u6c14\u7684\u751f\u7269\u5316\u5b66\u8fc7\u7a0b\u3002"<\/span>
\n partial =<\/span> functools.<\/span>partial(<\/span>eval<\/span>,<\/span> prompt =<\/span> CONCISENESS_PROMPT,<\/span> inputs =<\/span> inputs)<\/span>
\n await<\/span> partial(<\/span>outputs1)<\/span>
\n await<\/span> partial(<\/span>outputs2)<\/span><\/p>\n

asyncio.<\/span>run(<\/span>main(<\/span>)<\/span>)<\/span><\/p>\n

\u8f93\u51fa：<\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.15<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The response is highly verbose and includes multiple unnecessary elements: an opening compliment, background about plants, a long explanatory definition, a summary sentence, and a polite closing. While it does answer the question, it is far from the minimum concise form expected for a simple definition request. Thus, the score should be: 0.15."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.95<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output directly answers the question with a concise definition and includes only the essential information. It does not add hedging, pleasantries, meta-commentary, or unnecessary background beyond a minimal clarifying parenthetical example. Thus, the score should be: 0.95."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

2. \u6b63\u786e\u6027（CORRECTNESS）<\/h3>\n

\u6b63\u786e\u6027\u8bc4\u4f30\u7528\u4e8e\u786e\u5b9a\u8f93\u51fa\u5185\u5bb9\u7684\u4e8b\u5b9e\u51c6\u786e\u6027\u4ee5\u53ca\u5b8c\u6574\u6027\u3002\u5bf9\u4e8e\u4e00\u4e9b\u5e38\u8bc6\u6027\u6216\u8005\u516c\u5f00\u6027\u95ee\u9898，LLM\u662f\u77e5\u9053\u6b63\u786e\u7b54\u6848\u7684，\u4f46\u662f\u7531\u4e8e\u5982\u4e0b\u6240\u793a\u7684\u63d0\u793a\u8bcd\u6a21\u677f\u5b9a\u4e49\u4e86{inputs}\u3001{outputs}\u548c{reference_outputs}\u5360\u4f4d\u7b26，\u6240\u4ee5\u8f93\u5165\u548c\u5f15\u7528\u90fd\u9700\u8981\u6307\u5b9a，\u6b63\u786e\u6027\u8bc4\u4f30\u4f1a\u901a\u8fc7\u6bd4\u8f83\u8f93\u51fa\u548c\u5f15\u7528，\u5e76\u7ed3\u5408\u81ea\u5df1\u638c\u63e1\u7684\u77e5\u8bc6\u6765\u5224\u65ad\u5bf9\u9519\u3002\u6b63\u786e\u6027\u63d0\u793a\u8bcd\u5b9a\u4e49\u5728CORRECTNESS_PROMPT\u53d8\u91cf\u4e2d\u3002<\/p>\n

You are an expert data labeler evaluating model outputs for correctness. Your task is to assign a score based on the following rubric:<\/p>\n

<Rubric>
\n A correct answer:
\n – Provides accurate and complete information
\n – Contains no factual errors
\n – Addresses all parts of the question
\n – Is logically consistent
\n – Uses precise and accurate terminology<\/p>\n

When scoring, you should penalize:
\n – Factual errors or inaccuracies
\n – Incomplete or partial answers
\n – Misleading or ambiguous statements
\n – Incorrect terminology
\n – Logical inconsistencies
\n – Missing key information
\n<\/Rubric><\/p>\n

<Instructions>
\n – Carefully read the input and output
\n – Check for factual accuracy and completeness
\n – Focus on correctness of information rather than style or verbosity
\n<\/Instructions><\/p>\n

<Reminder>
\n The goal is to evaluate factual correctness and completeness of the response.
\n<\/Reminder><\/p>\n

<input>
\n{inputs}
\n<\/input><\/p>\n

<output>
\n{outputs}
\n<\/output><\/p>\n

Use the reference outputs below to help you evaluate the correctness of the response:<\/p>\n

<reference_outputs>
\n{reference_outputs}
\n<\/reference_outputs><\/p>\n

\u4ee5\u5982\u4e0b\u8fd9\u4e2a\u6f14\u793a\u7a0b\u5e8f\u4e3a\u4f8b，\u6211\u4eec\u91c7\u7528\u4e0a\u8ff0\u7684\u63d0\u793a\u8bcd，\u91c7\u7528\u4e0d\u540c\u7684reference_outputs\u5bf9\u540c\u4e00\u4e2a\u8f93\u51fa\u8fdb\u884c\u8bc4\u4f30\u3002\u5177\u4f53\u6765\u8bf4，\u5bf9\u4e8e\u95ee\u9898：\u6625\u79cb\u4e94\u9738\u6307\u7684\u662f\u90a3\u51e0\u4e2a?，\u6211\u4eec\u63d0\u4f9b\u4e86\u4e24\u4e2a\u6807\u51c6\u7b54\u6848（\u6625\u79cb\u4e94\u9738\u5386\u6765\u5c31\u6709\u4e24\u79cd\u8bf4\u6cd5）\u3002\u867d\u7136\u6587\u672c\u4e0d\u540c，\u5f88\u660e\u663e\u8f93\u5165\u4e0e\u7b2c\u4e00\u4e2a\u6807\u51c6\u7b54\u6848（\u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u548c\u5b8b\u8944\u516c）\u662f\u4e00\u81f4\u7684，\u8bc4\u4f30\u7684\u7ed3\u679c\u4e5f\u8bf4\u660e\u4e86\u8fd9\u4e00\u70b9\u3002<\/p>\n

async<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
\n inputs =<\/span>"\u6625\u79cb\u4e94\u9738\u662f\u90a3\u51e0\u4e2a？"<\/span>
\n outputs =<\/span> "\u9f50\u6853\u516c\u59dc\u5c0f\u767d\u3001\u664b\u6587\u516c\u59ec\u91cd\u8033\u3001\u695a\u5e84\u738b\u718a\u4fa3\u3001\u79e6\u7a46\u516c\u5b34\u4efb\u597d\u548c\u5b8b\u8944\u516c\u5b50\u5179\u752b"<\/span>
\n partial =<\/span> functools.<\/span>partial(<\/span>eval<\/span>,<\/span> prompt=<\/span>CORRECTNESS_PROMPT,<\/span> inputs=<\/span>inputs,<\/span> outputs =<\/span> outputs)<\/span>
\n reference_outputs1 =<\/span> "\u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u548c\u5b8b\u8944\u516c"<\/span>
\n reference_outputs2 =<\/span> "\u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u5434\u738b\u9616\u95fe\u3001\u8d8a\u738b\u52fe\u8df5"<\/span>
\n await<\/span> partial(<\/span>reference_outputs =<\/span> reference_outputs1)<\/span>
\n await<\/span> partial(<\/span>reference_outputs =<\/span> reference_outputs2)<\/span> <\/p>\n

\u8f93\u51fa：<\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 1.0<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output correctly identifies the traditional Spring and Autumn Five Hegemons: \u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u3001\u5b8b\u8944\u516c. It also gives their personal names, which are accurate in this context: \u59dc\u5c0f\u767d\u3001\u59ec\u91cd\u8033\u3001\u718a\u4fa3\u3001\u5b34\u4efb\u597d\u3001\u5b50\u5179\u752b. This matches the reference answer and fully addresses the question. Thus, the score should be: 1.0."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.4<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output lists the traditional \u201c\u4e94\u9738\u201d as \u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u3001\u5b8b\u8944\u516c, but the provided reference output gives the commonly accepted set as \u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u5434\u738b\u9616\u95fe\u3001\u8d8a\u738b\u52fe\u8df5. Therefore, the answer matches only the first three and gets the last two wrong, including an incorrect substitution of \u79e6\u7a46\u516c and \u5b8b\u8944\u516c for \u5434\u738b\u9616\u95fe and \u8d8a\u738b\u52fe\u8df5. It is incomplete relative to the reference and contains factual inaccuracies. Thus, the score should be: 0.4."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

\u4e3a\u4e86\u8bc1\u660eLLM\u5e76\u4e0d\u662f\u9488\u5bf9outputs\u548creference_outputs\u63d0\u4f9b\u7684\u6587\u672c\u8fdb\u884c\u63a8\u7406，\u8fd8\u4f1a\u5229\u7528\u81ea\u5df1\u638c\u63e1\u7684\u652f\u6301，\u6211\u4eec\u4fee\u6539\u4e86outputs\u53c2\u6570\u7684\u503c（\u59dc\u5c0f\u767d\u6362\u6210\u4e86\u59dc\u5c0f\u9ed1）\u3002\u4ece\u8f93\u51fa\u53ef\u4ee5\u770b\u51fa，\u8fd9\u79cd\u5c0f\u82b1\u62db\u7792\u4e0d\u8fc7LLM\u3002<\/p>\n

async<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
\n await<\/span> eval<\/span>(<\/span>
\n outputs =<\/span> "\u9f50\u6853\u516c\u59dc\u5c0f\u9ed1\u3001\u664b\u6587\u516c\u59ec\u91cd\u8033\u3001\u695a\u5e84\u738b\u718a\u4fa3\u3001\u79e6\u7a46\u516c\u5b34\u4efb\u597d\u548c\u5b8b\u8944\u516c\u5b50\u5179\u752b"<\/span>,<\/span>
\n prompt =<\/span> CORRECTNESS_PROMPT,<\/span>
\n inputs =<\/span> "\u6625\u79cb\u4e94\u9738\u662f\u90a3\u51e0\u4e2a？"<\/span>,<\/span>
\n reference_outputs =<\/span>"\u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u548c\u5b8b\u8944\u516c"<\/span>
\n )<\/span><\/p>\n

\u8f93\u51fa：<\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.9<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The response correctly identifies the traditional Spring and Autumn Five Hegemons: \u9f50\u6853\u516c\u3001\u664b\u6587\u516c\u3001\u695a\u5e84\u738b\u3001\u79e6\u7a46\u516c\u548c\u5b8b\u8944\u516c. It also includes their personal names, such as \u59dc\u5c0f\u767d for \u9f50\u6853\u516c, \u59ec\u91cd\u8033 for \u664b\u6587\u516c, \u718a\u4fa3 for \u695a\u5e84\u738b, \u5b34\u4efb\u597d for \u79e6\u7a46\u516c, and \u5b50\u5179\u752b for \u5b8b\u8944\u516c. However, there is a factual error: \u9f50\u6853\u516c\u2019s personal name is usually written \u59dc\u5c0f\u767d, not \u59dc\u5c0f\u9ed1. Because of this inaccuracy, the answer is not fully correct, though it is otherwise complete and relevant. Thus, the score should be: 0.9."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

3. \u56de\u7b54\u76f8\u5173\u6027（ANSWER_RELEVANCE）<\/h3>\n

\u56de\u7b54\u76f8\u5173\u6027\u8bc4\u4f30\u7528\u6765\u786e\u5b9a\u8f93\u51fa\u662f\u5426\u76f4\u63a5\u56de\u7b54\u4e86\u7528\u6237\u6240\u63d0\u51fa\u7684\u95ee\u9898，\u7528\u6765\u89e3\u51b3\u9632\u7b54\u975e\u6240\u95ee\u7684\u95ee\u9898\u2014\u2014\u6709\u65f6\u5019\u6a21\u578b\u56de\u7b54\u5f97\u5f88\u5bf9，\u4f46\u6839\u672c\u4e0d\u662f\u7528\u6237\u95ee\u7684\u95ee\u9898（\u6bd4\u5982\u95ee\u57ce\u95e8\u697c\u5b50，\u7b54\u80ef\u9aa8\u8f74\u5b50）\u3002\u5bf9\u5e94\u63d0\u793a\u8bcd\u6587\u672c\u5185\u5bb9\u5982\u4e0b，\u5bf9\u5e94\u7684\u53d8\u91cf\u4e3aANSWER_RELEVANCE_PROMPT\u3002\u7531\u4e8e\u6a21\u677f\u5305\u542b{inputs}\u548c{outputs}，\u6240\u4ee5\u5728\u6267\u884c\u8bc4\u4f30\u5668\u7684\u65f6\u5019\u9700\u8981\u6307\u5b9a\u8f93\u5165\u3002<\/p>\n

You are an expert evaluator assessing whether outputs are relevant to the given input. Your task is to determine whether EACH statement appropriately addresses what was asked.<\/p>\n

<Rubric>
\nA relevant output:
\n– Directly answers the question or addresses the request
\n– Provides information specifically asked for
\n– Stays on topic with the input's intent
\n– Contributes meaningfully to fulfilling the request<\/p>\n

An irrelevant output:
\n– Discusses topics not requested or implied by the input
\n– Provides unnecessary tangents or digressions
\n– Includes information that doesn't answer the question
\n– Addresses a different question than what was asked
\n<\/Rubric><\/p>\n

<Instructions>
\nFor each output:
\n– Read the original input carefully to understand what was asked
\n– Examine the output and identify its core claim or purpose
\n– Determine if the output directly addresses the input's request
\n– Assess whether the information helps fulfill what was asked
\n– Determine the answer relevancy of output and output a score
\n<\/Instructions><\/p>\n

<Reminder>
\nFocus on whether each statement helps answer the specific input question, not whether the statement is true or well-written. A statement can be factually correct but still irrelevant if it doesn't address what was asked.
\n<\/Reminder><\/p>\n

Now, grade the following example according to the above instructions:<\/p>\n

<example>
\n<input>
\n{inputs}
\n<\/input><\/p>\n

<output>
\n{outputs}
\n<\/output>
\n<\/example><\/p>\n

\u5728\u5982\u4e0b\u7684\u6f14\u793a\u7a0b\u5e8f\u4e2d，\u6211\u4eec\u4f7f\u7528\u4e0a\u8ff0\u7684\u9884\u5b9a\u4e49\u63d0\u793a\u8bcd\u5b9e\u65bd\u76f8\u5173\u6027\u8bc4\u4f30\u3002\u9488\u5bf9\u8f93\u5165\u7684\u95ee\u9898\u4eca\u5929\u4e0b\u5348\u5317\u4eac\u4f1a\u4e0b\u96e8\u5417？\u6211\u9700\u8981\u5e26\u4f1e\u5417？，\u6211\u4eec\u63d0\u4f9b\u4e86\u4e24\u4e2a\u7b54\u6848\u3002\u524d\u8005\u5f15\u5165\u5f88\u591a\u65e0\u5173\u4fe1\u606f，\u5f3a\u884c\u626f\u5230\u6e29\u5e26\u5b63\u98ce\u6c14\u5019\u548c\u9632\u6652\u4fdd\u62a4\u76ae\u80a4，\u4e25\u91cd\u504f\u79bb\u4e86\u7528\u6237\u5373\u65f6\u6027\u7684\u67e5\u8be2\u9700\u6c42；\u540e\u8005\u5219\u4e0e\u95ee\u9898\u9ad8\u5ea6\u76f8\u5173\u3002\u8f93\u51fa\u7684\u8bc4\u4f30\u7ed3\u679c\u4e5f\u4f53\u73b0\u4e86\u5b83\u4eec\u4e0e\u95ee\u9898\u7684\u76f8\u5173\u7a0b\u5ea6\u3002<\/p>\n

async<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
\n partial =<\/span> functools.<\/span>partial(<\/span>
\n eval<\/span>,<\/span>
\n prompt =<\/span> ANSWER_RELEVANCE_PROMPT,<\/span>
\n inputs =<\/span> "\u4eca\u5929\u4e0b\u5348\u5317\u4eac\u4f1a\u4e0b\u96e8\u5417？\u6211\u9700\u8981\u5e26\u4f1e\u5417？"<\/span>
\n )<\/span><\/p>\n

outputs1 =<\/span> """\\\\
\n\u4e0b\u96e8\u662f\u4e00\u79cd\u5e38\u89c1\u7684\u81ea\u7136\u964d\u6c34\u73b0\u8c61\u3002
\n\u5317\u4eac\u4f5c\u4e3a\u4e2d\u56fd\u7684\u9996\u90fd，\u5c5e\u4e8e\u6e29\u5e26\u5b63\u98ce\u6c14\u5019，\u590f\u5b63\u964d\u6c34\u975e\u5e38\u96c6\u4e2d，\u7ecf\u5e38\u4f1a\u51fa\u73b0\u5f3a\u5bf9\u6d41\u5929\u6c14\u548c\u96f7\u9635\u96e8\u3002
\n\u5982\u679c\u60a8\u5728\u590f\u5b63\u51fa\u884c，\u5efa\u8bae\u60a8\u968f\u65f6\u5173\u6ce8\u6c14\u8c61\u53f0\u53d1\u5e03\u7684\u6700\u65b0\u9884\u62a5，\u4ee5\u4fbf\u5408\u7406\u5b89\u6392\u884c\u7a0b，\u907f\u514d\u56e0\u6076\u52a3\u5929\u6c14\u803d\u8bef\u60a8\u7684\u65f6\u95f4\u3002
\n\u53e6\u5916，\u51fa\u95e8\u5e26\u4e00\u628a\u6674\u96e8\u4f1e\u4e0d\u4ec5\u80fd\u9632\u96e8，\u5728\u70c8\u65e5\u4e0b\u8fd8\u80fd\u8d77\u5230\u9632\u6652\u7684\u4f5c\u7528，\u5bf9\u4fdd\u62a4\u76ae\u80a4\u5f88\u6709\u5e2e\u52a9\u3002"""<\/span><\/p>\n

outputs2 =<\/span> """\\\\
\n\u6839\u636e\u6700\u65b0\u7684\u6c14\u8c61\u9884\u62a5，\u4eca\u5929\u4e0b\u5348\u5317\u4eac\u5927\u90e8\u5206\u5730\u533a\u4e3a\u591a\u4e91\u8f6c\u9634，\u508d\u665a\u524d\u540e\u6709 80% \u7684\u6982\u7387\u51fa\u73b0\u96f7\u9635\u96e8\u3002
\n\u56e0\u6b64，\u5f3a\u70c8\u5efa\u8bae\u60a8\u51fa\u95e8\u5e26\u4f1e，\u65e2\u80fd\u9632\u96e8\u4e5f\u80fd\u9632\u8303\u7a81\u53d1\u7684\u5f3a\u5bf9\u6d41\u5929\u6c14\u3002"""<\/span><\/p>\n

await<\/span> partial(<\/span>outputs1)<\/span>
\n await<\/span> partial(<\/span>outputs2)<\/span><\/p>\n

\u8f93\u51fa：<\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.4<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The input asks a specific, time-sensitive question: whether it will rain in Beijing this afternoon and whether to bring an umbrella. The output gives general information about rain, Beijing climate, and generic advice to check forecasts and carry an umbrella, but it does not directly answer the actual weather question or clearly determine whether an umbrella is needed for this afternoon. It is related in theme, but it only partially addresses the request and lacks the specific answer asked for. Thus, the score should be: 0.4."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 1.0<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output directly addresses the user's question about whether it will rain in Beijing this afternoon and whether they need to bring an umbrella. It provides a weather forecast and a clear recommendation to bring an umbrella, which is exactly what was asked. Thus, the score should be: 1.0."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

4. \u5e7b\u89c9\u5ea6（HALLUCINATION）<\/h3>\n

\u5e7b\u89c9\u5ea6\u8bc4\u4f30\u7528\u4e8e\u786e\u5b9a\u8f93\u51fa\u4e2d\u662f\u5426\u5305\u542b\u4e86\u65e0\u6cd5\u88ab\u6240\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u652f\u6301\u7684\u4fe1\u606f\u3002\u8fd9\u79cd\u8bc4\u4f30\u5728RAG\u4e2d\u5177\u6709\u5e7f\u6cdb\u7684\u5e94\u7528\u2014\u2014\u68c0\u67e5\u6a21\u578b\u6709\u6ca1\u6709\u778e\u7f16\u3002\u5982\u679c\u53c2\u8003\u6587\u6863\u91cc\u6ca1\u5199，\u6a21\u578b\u81ea\u5df1\u8111\u8865\u4e86，\u5c31\u4f1a\u88ab\u5224\u5b9a\u4e3a\u5e7b\u89c9\u3002\u5e7b\u89c9\u5ea6\u8bc4\u4f30\u63d0\u793a\u8bcd\u5b9a\u4e49\u5728\u540d\u4e3aHALLUCINATION_PROMPT\u7684\u53d8\u91cf\u4e2d，\u5177\u4f53\u5185\u5bb9\u5982\u4e0b\u3002\u6a21\u677f\u6587\u672c\u4e2d\u5305\u542b{context}\u3001{inputs}\u3001{outputs}\u548c{reference_outputs}\u56db\u4e2a\u5360\u4f4d\u7b26，\u8c03\u7528\u5bf9\u5e94\u7684\u8bc4\u4f30\u5668\u65f6\u5fc5\u987b\u6307\u5b9a\u3002<\/p>\n

You are an expert data labeler evaluating model outputs for hallucinations. Your task is to assign a score based on the following rubric:<\/p>\n

<Rubric>
\n A response without hallucinations:
\n – Contains only verifiable facts that are directly supported by the input context
\n – Makes no unsupported claims or assumptions
\n – Does not add speculative or imagined details
\n – Maintains perfect accuracy in dates, numbers, and specific details
\n – Appropriately indicates uncertainty when information is incomplete
\n<\/Rubric><\/p>\n

<Instructions>
\n – Read the input context thoroughly
\n – Identify all claims made in the output
\n – Cross-reference each claim with the input context
\n – Note any unsupported or contradictory information
\n – Consider the severity and quantity of hallucinations
\n<\/Instructions><\/p>\n

<Reminder>
\n Focus solely on factual accuracy and support from the input context. Do not consider style, grammar, or presentation in scoring. A shorter, factual response should score higher than a longer response with unsupported claims.
\n<\/Reminder><\/p>\n

Use the following context to help you evaluate for hallucinations in the output:<\/p>\n

<context>
\n{context}
\n<\/context><\/p>\n

<input>
\n{inputs}
\n<\/input><\/p>\n

<output>
\n{outputs}
\n<\/output><\/p>\n

If available, you may also use the reference outputs below to help you identify hallucinations in the response:<\/p>\n

<reference_outputs>
\n{reference_outputs}
\n<\/reference_outputs><\/p>\n

\u5728\u4e0b\u9762\u7684\u6f14\u793a\u7a0b\u5e8f\u4e2d，\u6211\u4eec\u4f7f\u7528\u4e0a\u8ff0\u7684\u63d0\u793a\u8bcd\u9488\u5bf9\u95ee\u9898**\u7279\u65af\u62c9\u516c\u53f8\u662f\u5728\u54ea\u4e00\u5e74\u7531\u8c01\u521b\u7acb\u7684？**\u7684\u4e24\u4e2a\u7b54\u6848\u5b9e\u65bd\u5e7b\u89c9\u8bc4\u4f30\u3002\u6839\u636e\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587，\u7b2c\u4e00\u4e2a\u56de\u7b54（\u7279\u65af\u62c9\u7531\u9a6c\u4e01\u00b7\u57c3\u4f2f\u54c8\u5fb7\u548c\u9a6c\u514b\u00b7\u5854\u5f6d\u5b81\u4e8e2003\u5e74\u521b\u7acb\u3002）\u624d\u662f\u6b63\u786e\u7684，\u8fd9\u4e5f\u5728\u8f93\u51fa\u7684\u8bc4\u4f30\u7ed3\u679c\u7684\u5f97\u5230\u5370\u8bc1\u3002<\/p>\n

async<\/span> def<\/span> main<\/span>(<\/span>)<\/span>:<\/span>
\n partial =<\/span> functools.<\/span>partial(<\/span>
\n eval<\/span>,<\/span>
\n prompt =<\/span> HALLUCINATION_PROMPT,<\/span>
\n inputs =<\/span> "\u7279\u65af\u62c9\u516c\u53f8\u662f\u5728\u54ea\u4e00\u5e74\u7531\u8c01\u521b\u7acb\u7684？"<\/span>,<\/span>
\n reference_outputs =<\/span>"\u7279\u65af\u62c9\u7531\u9a6c\u4e01\u00b7\u57c3\u4f2f\u54c8\u5fb7\u548c\u9a6c\u514b\u00b7\u5854\u5f6d\u5b81\u4e8e2003\u5e74\u521b\u7acb\u3002"<\/span>,<\/span>
\n context =<\/span> """\\\\
\n\u7279\u65af\u62c9（Tesla）\u662f\u4e00\u5bb6\u7f8e\u56fd\u7535\u52a8\u6c7d\u8f66\u53ca\u80fd\u6e90\u516c\u53f8\u3002
\n\u8be5\u516c\u53f8\u7531\u9a6c\u4e01\u00b7\u57c3\u4f2f\u54c8\u5fb7（Martin Eberhard）\u548c\u9a6c\u514b\u00b7\u5854\u5f6d\u5b81（Marc Tarpenning）\u4e8e2003\u5e747\u67081\u65e5\u521b\u7acb，\u603b\u90e8\u6700\u521d\u4f4d\u4e8e\u7f8e\u56fd\u52a0\u5229\u798f\u5c3c\u4e9a\u5dde\u3002
\n\u516c\u53f8\u7684\u540d\u5b57\u662f\u4e3a\u4e86\u7eaa\u5ff5\u7269\u7406\u5b66\u5bb6\u5c3c\u53e4\u62c9\u00b7\u7279\u65af\u62c9\u3002"""<\/span>)<\/span><\/p>\n

outputs1 =<\/span> "\u7279\u65af\u62c9\u7531\u9a6c\u4e01\u00b7\u57c3\u4f2f\u54c8\u5fb7\u548c\u9a6c\u514b\u00b7\u5854\u5f6d\u5b81\u4e8e2003\u5e74\u521b\u7acb\u3002"<\/span>
\n outputs2 =<\/span> "\u7279\u65af\u62c9\u516c\u53f8\u662f\u7531\u8457\u540d\u4f01\u4e1a\u5bb6\u57c3\u9686\u00b7\u9a6c\u65af\u514b（Elon Musk）\u4e8e2003\u5e747\u67081\u65e5\u521b\u7acb\u7684\u3002"<\/span>
\n await<\/span> partial(<\/span>outputs1)<\/span>
\n await<\/span> partial(<\/span>outputs2)<\/span><\/p>\n

\u8f93\u51fa：<\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 1.0<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output states that Tesla was founded by Martin Eberhard and Marc Tarpenning in 2003, which is fully supported by the provided context and reference output. It does not add any unsupported details or contradict the source, though it omits the exact founding date and does not need to because the question asks for the year and founders. Thus, the score should be: 1.0."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n

{<\/span>
\n "key"<\/span>:<\/span> "score"<\/span>,<\/span>
\n "score"<\/span>:<\/span> 0.2<\/span>,<\/span>
\n "comment"<\/span>:<\/span> "The output incorrectly states that Tesla was founded by Elon Musk. The provided context says Tesla was founded by Martin Eberhard and Marc Tarpenning on July 1, 2003, so the founder claim is unsupported and contradictory. It also says \\\\"\u8457\u540d\u4f01\u4e1a\u5bb6\\\\" (famous entrepreneur), which is not supported by the context. The year and date are partly supported, but the key factual claim about who founded the company is wrong. Thus, the score should be: 0.2."<\/span>,<\/span>
\n "metadata"<\/span>:<\/span> null<\/span>
\n}<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

\u5bf9\u4e8e\u6211\u4eec\u4e2a\u4eba\u6765\u8bf4，Agent\u7f16\u7a0b\u6700\u5927\u7684\u6311\u6218\u4e0d\u662f\u5404\u79cd\u6240\u8c13\u7684\u5de5\u4f5c，\u800c\u662f\u5199\u63d0\u793a\u8bcd，\u6bd5\u7adf\u6211\u4e0d\u662f\u4e13\u4e1a\u7684\u6587\u5b57\u5de5\u4f5c\u8005\u3002OpenEvals\u9488\u5bf9\u4e00\u7cfb\u5217\u9488\u5bf9\u8f93\u51fa\u6587\u672c\u8d28\u91cf\u7684\u8bc4\u4f30\u6307\u6807\u5b9a\u4e49\u4e86\u5bf9\u5e94\u7684\u63d0\u793a\u8bcd，\u7ed3\u5408\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-02：LLM-as-a-Judge:\u8ba9LLM\u5f53\u88c1\u5224\u6765\u8bc4\u4f30Agent\u7684\u8f93\u51fa\u4ecb\u7ecd\u4e86\u57fa\u4e8eLLM-as-a-Judge\u7684\u8bc4\u4f30\u5668，\u6211\u4eec\u53ef\u4ee5\u8fdb\u884c\u9488\u5bf9\u7b80\u6d01\u5ea6\u3001\u6b63\u786e\u6027\u3001\u5e7b\u89c9\u5ea6\u3001\u76f8\u5173\u6027\u3001\u8ba1\u5212\u9075\u5faa\u5ea6<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[10480,10481,10336,74,66,997],"topic":[],"class_list":["post-91840","post","type-post","status-publish","format-standard","hentry","category-server","tag-evaluation","tag-openevals","tag-10336","tag-agent","tag-ai","tag-langchain"],"yoast_head":"\n[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/91840.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u5bf9\u4e8e\u6211\u4eec\u4e2a\u4eba\u6765\u8bf4，Agent\u7f16\u7a0b\u6700\u5927\u7684\u6311\u6218\u4e0d\u662f\u5404\u79cd\u6240\u8c13\u7684\u5de5\u4f5c，\u800c\u662f\u5199\u63d0\u793a\u8bcd，\u6bd5\u7adf\u6211\u4e0d\u662f\u4e13\u4e1a\u7684\u6587\u5b57\u5de5\u4f5c\u8005\u3002OpenEvals\u9488\u5bf9\u4e00\u7cfb\u5217\u9488\u5bf9\u8f93\u51fa\u6587\u672c\u8d28\u91cf\u7684\u8bc4\u4f30\u6307\u6807\u5b9a\u4e49\u4e86\u5bf9\u5e94\u7684\u63d0\u793a\u8bcd，\u7ed3\u5408\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-02：LLM-as-a-Judge:\u8ba9LLM\u5f53\u88c1\u5224\u6765\u8bc4\u4f30Agent\u7684\u8f93\u51fa\u4ecb\u7ecd\u4e86\u57fa\u4e8eLLM-as-a-Judge\u7684\u8bc4\u4f30\u5668，\u6211\u4eec\u53ef\u4ee5\u8fdb\u884c\u9488\u5bf9\u7b80\u6d01\u5ea6\u3001\u6b63\u786e\u6027\u3001\u5e7b\u89c9\u5ea6\u3001\u76f8\u5173\u6027\u3001\u8ba1\u5212\u9075\u5faa\u5ea6\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/91840.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-08T07:00:52+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/91840.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/91840.html\",\"name\":\"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-08T07:00:52+00:00\",\"dateModified\":\"2026-08-08T07:00:52+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/91840.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/91840.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/91840.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7]\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/91840.html","og_locale":"zh_CN","og_type":"article","og_title":"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u5bf9\u4e8e\u6211\u4eec\u4e2a\u4eba\u6765\u8bf4，Agent\u7f16\u7a0b\u6700\u5927\u7684\u6311\u6218\u4e0d\u662f\u5404\u79cd\u6240\u8c13\u7684\u5de5\u4f5c，\u800c\u662f\u5199\u63d0\u793a\u8bcd，\u6bd5\u7adf\u6211\u4e0d\u662f\u4e13\u4e1a\u7684\u6587\u5b57\u5de5\u4f5c\u8005\u3002OpenEvals\u9488\u5bf9\u4e00\u7cfb\u5217\u9488\u5bf9\u8f93\u51fa\u6587\u672c\u8d28\u91cf\u7684\u8bc4\u4f30\u6307\u6807\u5b9a\u4e49\u4e86\u5bf9\u5e94\u7684\u63d0\u793a\u8bcd，\u7ed3\u5408\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-02：LLM-as-a-Judge:\u8ba9LLM\u5f53\u88c1\u5224\u6765\u8bc4\u4f30Agent\u7684\u8f93\u51fa\u4ecb\u7ecd\u4e86\u57fa\u4e8eLLM-as-a-Judge\u7684\u8bc4\u4f30\u5668，\u6211\u4eec\u53ef\u4ee5\u8fdb\u884c\u9488\u5bf9\u7b80\u6d01\u5ea6\u3001\u6b63\u786e\u6027\u3001\u5e7b\u89c9\u5ea6\u3001\u76f8\u5173\u6027\u3001\u8ba1\u5212\u9075\u5faa\u5ea6","og_url":"https:\/\/www.wsisp.com\/helps\/91840.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-08T07:00:52+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"10 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/91840.html","url":"https:\/\/www.wsisp.com\/helps\/91840.html","name":"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7] - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-08T07:00:52+00:00","dateModified":"2026-08-08T07:00:52+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/91840.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/91840.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/91840.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"[\u57fa\u4e8eOpenEvals\u7684\u81ea\u52a8\u5316\u8bc4\u4f30-06]\u8bc4\u4f30Agent\u8f93\u51fa\u6587\u672c\u7684\u8d28\u91cf[\u4e0a\u7bc7]"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/91840","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=91840"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/91840\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=91840"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=91840"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=91840"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=91840"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}