{"id":34618,"date":"2025-08-01T12:09:14","date_gmt":"2025-08-01T04:09:14","guid":{"rendered":"https:\/\/www.kdjingpai.com\/?p=34618"},"modified":"2025-08-01T12:09:14","modified_gmt":"2025-08-01T04:09:14","slug":"openbench","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/openbench\/","title":{"rendered":"OpenBench\uff1a\u4e00\u4e2a\u7528\u4e8e\u8bc4\u4f30\u8bed\u8a00\u6a21\u578b\u7684\u5f00\u6e90\u57fa\u51c6\u6d4b\u8bd5\u5de5\u5177"},"content":{"rendered":"<p>OpenBench\u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u8bed\u8a00\u6a21\u578b\u8bc4\u4f30\u5de5\u5177\uff0c\u5b83\u4e0d\u53d7\u9650\u4e8e\u7279\u5b9a\u7684\u6a21\u578b\u4f9b\u5e94\u5546\u3002\u5f00\u53d1\u8005\u53ef\u4ee5\u4f7f\u7528\u5b83\u5728\u8d85\u8fc720\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u96c6\u4e0a\u5bf9\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u6807\u51c6\u5316\u7684\u3001\u53ef\u590d\u73b0\u7684\u6027\u80fd\u8bc4\u6d4b\uff0c\u8fd9\u4e9b\u6d4b\u8bd5\u6db5\u76d6\u4e86\u77e5\u8bc6\u3001\u63a8\u7406\u3001\u7f16\u7801\u548c\u6570\u5b66\u7b49\u591a\u4e2a\u9886\u57df\u3002OpenBench\u7684\u6838\u5fc3\u4f18\u52bf\u5728\u4e8e\u5176\u7b80\u6d01\u6027\u548c\u901a\u7528\u6027\u3002\u5b83\u63d0\u4f9b\u4e86\u4e00\u4e2a\u7b80\u5355\u7684\u547d\u4ee4\u884c\u754c\u9762\uff0c\u7528\u6237\u53ea\u9700\u51e0\u6761\u547d\u4ee4\u5c31\u53ef\u4ee5\u542f\u52a8\u8bc4\u4f30\u4efb\u52a1\u3002\u8be5\u5de5\u5177\u652f\u6301\u591a\u79cd\u4e3b\u6d41\u6a21\u578b\u670d\u52a1\u5546\uff0c\u4f8b\u5982Groq\u3001OpenAI\u3001Anthropic\u3001\u8c37\u6b4c\uff0c\u540c\u65f6\u4e5f\u517c\u5bb9\u901a\u8fc7Ollama\u8fd0\u884c\u7684\u672c\u5730\u6a21\u578b\u3002\u7531\u4e8e\u5b83\u6784\u5efa\u4e8e<code>inspect-ai<\/code>\u6846\u67b6\u4e4b\u4e0a\uff0c\u56e0\u6b64\u5177\u5907\u826f\u597d\u7684\u6269\u5c55\u6027\uff0c\u5141\u8bb8\u5f00\u53d1\u8005\u65b9\u4fbf\u5730\u6dfb\u52a0\u65b0\u7684\u57fa\u51c6\u6d4b\u8bd5\u548c\u8bc4\u4f30\u6307\u6807\u3002\u8fd9\u4f7f\u5f97OpenBench\u6210\u4e3a\u4e00\u4e2a\u7075\u6d3b\u4e14\u6613\u4e8e\u4e0a\u624b\u7684\u6a21\u578b\u6027\u80fd\u8bc4\u6d4b\u5e73\u53f0\u3002<\/p>\n<h2>\u529f\u80fd\u5217\u8868<\/h2>\n<ul>\n<li><strong>\u652f\u6301\u8d85\u8fc720\u4e2a\u57fa\u51c6\u6d4b\u8bd5<\/strong>\uff1a\u5185\u7f6eMMLU\u3001GPQA\u3001HumanEval\u3001SimpleQA\u4ee5\u53caAIME\u3001HMMT\u7b49\u591a\u79cd\u7ade\u8d5b\u7ea7\u6570\u5b66\u8bc4\u6d4b\u3002<\/li>\n<li><strong>\u7b80\u6d01\u7684\u547d\u4ee4\u884c\u754c\u9762(CLI)<\/strong>\uff1a\u63d0\u4f9b<code>bench list<\/code>\u3001<code>bench describe<\/code>\u3001<code>bench eval<\/code>\u7b49\u7b80\u5355\u76f4\u89c2\u7684\u547d\u4ee4\u6765\u7ba1\u7406\u548c\u8fd0\u884c\u8bc4\u6d4b\u3002<\/li>\n<li><strong>\u517c\u5bb9\u591a\u5bb6\u6a21\u578b\u4f9b\u5e94\u5546<\/strong>\uff1a\u652f\u6301\u8d85\u8fc715\u5bb6\u6a21\u578b\u4f9b\u5e94\u5546\uff0c\u5305\u62ecGroq\u3001OpenAI\u3001Anthropic\u3001Google\u3001AWS Bedrock\u3001Azure\u7b49\u3002<\/li>\n<li><strong>\u652f\u6301\u672c\u5730\u6a21\u578b<\/strong>\uff1a\u53ef\u4ee5\u4e0eOllama\u96c6\u6210\uff0c\u5bf9\u5728\u672c\u5730\u8fd0\u884c\u7684\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u8bc4\u4f30\u3002<\/li>\n<li><strong>\u57fa\u4e8e\u6807\u51c6\u6846\u67b6\u6784\u5efa<\/strong>\uff1a\u6784\u5efa\u5728<code>inspect-ai<\/code>\u8bc4\u4f30\u6846\u67b6\u4e4b\u4e0a\uff0c\u786e\u4fdd\u4e86\u8bc4\u6d4b\u7684\u89c4\u8303\u6027\u548c\u53ef\u9760\u6027\u3002<\/li>\n<li><strong>\u9ad8\u5ea6\u53ef\u6269\u5c55<\/strong>\uff1a\u5141\u8bb8\u5f00\u53d1\u8005\u8f7b\u677e\u6dfb\u52a0\u65b0\u7684\u57fa\u51c6\u6d4b\u8bd5\u9879\u76ee\u548c\u81ea\u5b9a\u4e49\u8bc4\u4f30\u6307\u6807\u3002<\/li>\n<li><strong>\u4ea4\u4e92\u5f0f\u7ed3\u679c\u67e5\u770b<\/strong>\uff1a\u63d0\u4f9b<code>bench view<\/code>\u547d\u4ee4\uff0c\u53ef\u4ee5\u5728\u4ea4\u4e92\u5f0f\u7528\u6237\u754c\u9762\u4e2d\u67e5\u770b\u8bc4\u4f30\u65e5\u5fd7\u3002<\/li>\n<li><strong>\u7075\u6d3b\u7684\u8bc4\u4f30\u914d\u7f6e<\/strong>\uff1a\u7528\u6237\u53ef\u4ee5\u901a\u8fc7\u547d\u4ee4\u884c\u53c2\u6570\u6216\u73af\u5883\u53d8\u91cf\u8be6\u7ec6\u914d\u7f6e\u6a21\u578b\u8bc4\u4f30\u8fc7\u7a0b\uff0c\u5982\u8bbe\u7f6e\u6e29\u5ea6\u3001\u6700\u5927Token\u6570\u3001\u5e76\u53d1\u8bf7\u6c42\u6570\u7b49\u3002<\/li>\n<\/ul>\n<h2>\u4f7f\u7528\u5e2e\u52a9<\/h2>\n<p>OpenBench \u63d0\u4f9b\u4e86\u4e00\u5957\u5b8c\u6574\u7684\u5de5\u5177\uff0c\u7528\u4e8e\u5bf9\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u8fdb\u884c\u6807\u51c6\u5316\u7684\u57fa\u51c6\u6d4b\u8bd5\u3002\u4e0b\u9762\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u5982\u4f55\u5b89\u88c5\u548c\u4f7f\u7528\u8be5\u5de5\u5177\u6765\u8bc4\u4f30\u6a21\u578b\u3002<\/p>\n<h3><strong>1. \u73af\u5883\u51c6\u5907\u4e0e\u5b89\u88c5<\/strong><\/h3>\n<p>\u5728\u4f7f\u7528 OpenBench \u4e4b\u524d\uff0c\u4f60\u9700\u8981\u5148\u5b89\u88c5\u00a0<code>uv<\/code>\uff0c\u5b83\u662f\u4e00\u4e2a\u5feb\u901f\u7684 Python \u5305\u5b89\u88c5\u5668\u548c\u865a\u62df\u73af\u5883\u7ba1\u7406\u5668\u3002<\/p>\n<p><strong>\u6b65\u9aa4\u4e00\uff1a\u5b89\u88c5 uv (\u5982\u679c\u5c1a\u672a\u5b89\u88c5)<\/strong><br \/>\n<code>uv<\/code>\u00a0\u7684\u5b89\u88c5\u8fc7\u7a0b\u975e\u5e38\u7b80\u5355\uff0c\u5177\u4f53\u53ef\u4ee5\u53c2\u8003\u5176\u5b98\u65b9\u6587\u6863\u3002\u5b89\u88c5\u5b8c\u6210\u540e\uff0c\u4f60\u5c31\u53ef\u4ee5\u5f00\u59cb\u51c6\u5907 OpenBench \u7684\u8fd0\u884c\u73af\u5883\u3002<\/p>\n<p><strong>\u6b65\u9aa4\u4e8c\uff1a\u521b\u5efa\u5e76\u6fc0\u6d3b\u865a\u62df\u73af\u5883<\/strong><br \/>\n\u4e3a\u4e86\u4fdd\u6301\u9879\u76ee\u4f9d\u8d56\u7684\u9694\u79bb\uff0c\u5efa\u8bae\u521b\u5efa\u4e00\u4e2a\u65b0\u7684\u865a\u62df\u73af\u5883\u3002<\/p>\n<pre><code># \u521b\u5efa\u4e00\u4e2a\u540d\u4e3a .venv \u7684\u865a\u62df\u73af\u5883\r\nuv venv\r\n# \u6fc0\u6d3b\u8be5\u865a\u62df\u73af\u5883 (\u5728Linux\u6216macOS\u4e0a)\r\nsource .venv\/bin\/activate\r\n<\/code><\/pre>\n<p><strong>\u6b65\u9aa4\u4e09\uff1a\u5b89\u88c5 OpenBench<\/strong><br \/>\n\u6fc0\u6d3b\u865a\u62df\u73af\u5883\u540e\uff0c\u4f7f\u7528\u00a0<code>uv<\/code>\u00a0\u6765\u5b89\u88c5 OpenBench\u3002<\/p>\n<pre><code>uv pip install openbench```\r\n\u8fd9\u4e2a\u547d\u4ee4\u4f1a\u81ea\u52a8\u5904\u7406\u6240\u6709\u5fc5\u9700\u7684\u4f9d\u8d56\u9879\u3002\r\n### **2. \u914d\u7f6e API \u5bc6\u94a5**\r\nOpenBench \u652f\u6301\u591a\u5bb6\u6a21\u578b\u4f9b\u5e94\u5546\uff0c\u4f60\u9700\u8981\u8bbe\u7f6e\u76f8\u5e94\u7684 API \u5bc6\u94a5\u624d\u80fd\u4f7f\u7528\u5b83\u4eec\u7684\u6a21\u578b\u3002\u5bc6\u94a5\u901a\u8fc7\u73af\u5883\u53d8\u91cf\u8fdb\u884c\u914d\u7f6e\u3002\r\n```bash\r\n# \u793a\u4f8b\uff1a\u8bbe\u7f6e <a href=\"https:\/\/www.kdjingpai.com\/groqaidamoxingbai\/\">Groq<\/a> \u7684 API \u5bc6\u94a5\r\nexport GROQ_API_KEY=\"\u4f60\u7684\u5bc6\u94a5\"\r\n# \u793a\u4f8b\uff1a\u8bbe\u7f6e OpenAI \u7684 API \u5bc6\u94a5\r\nexport OPENAI_API_KEY=\"\u4f60\u7684\u5bc6\u94a5\"\r\n# \u793a\u4f8b\uff1a\u8bbe\u7f6e <a href=\"https:\/\/www.kdjingpai.com\/claudeanquanfubai\/\">Anthropic<\/a> \u7684 API \u5bc6\u94a5\r\nexport ANTHROPIC_API_KEY=\"\u4f60\u7684\u5bc6\u94a5\"```\r\n\u4f60\u53ea\u9700\u8981\u8bbe\u7f6e\u4f60\u8ba1\u5212\u4f7f\u7528\u7684\u90a3\u4e2a\u4f9b\u5e94\u5546\u7684\u5bc6\u94a5\u5373\u53ef\u3002\r\n### **3. \u8fd0\u884c\u8bc4\u4f30\u4efb\u52a1**\r\n\u914d\u7f6e\u5b8c\u6210\u540e\uff0c\u4f60\u5c31\u53ef\u4ee5\u901a\u8fc7 `bench eval` \u547d\u4ee4\u6765\u8fd0\u884c\u4e00\u4e2a\u8bc4\u4f30\u4efb\u52a1\u3002\r\n**\u57fa\u672c\u547d\u4ee4\u683c\u5f0f\uff1a**\r\n`bench eval &lt;\u57fa\u51c6\u6d4b\u8bd5\u540d\u79f0&gt; --model &lt;\u6a21\u578b\u540d\u79f0&gt;`\r\n**\u5feb\u901f\u4e0a\u624b\u793a\u4f8b\uff1a**\r\n\u8ba9\u6211\u4eec\u4ee5`mmlu`\u57fa\u51c6\u6d4b\u8bd5\u4e3a\u4f8b\uff0c\u4f7f\u7528Groq\u63d0\u4f9b\u7684`llama-3.3-70b-versatile`\u6a21\u578b\uff0c\u5e76\u53ea\u8bc4\u4f3010\u4e2a\u6837\u672c\u3002\r\n```bash\r\nbench eval mmlu --model groq\/llama-3.3-70b-versatile --limit 10\r\n<\/code><\/pre>\n<ul>\n<li><code>mmlu<\/code>: \u8fd9\u662f\u57fa\u51c6\u6d4b\u8bd5\u7684\u540d\u79f0\u3002<\/li>\n<li><code>--model groq\/llama-3.3-70b-versatile<\/code>: \u6307\u5b9a\u8981\u8bc4\u4f30\u7684\u6a21\u578b\u3002<\/li>\n<li><code>--limit 10<\/code>: \u8868\u793a\u53ea\u4ece\u6570\u636e\u96c6\u4e2d\u9009\u53d610\u4e2a\u6837\u672c\u8fdb\u884c\u6d4b\u8bd5\uff0c\u8fd9\u5728\u521d\u6b21\u5c1d\u8bd5\u65f6\u53ef\u4ee5\u5feb\u901f\u5f97\u5230\u7ed3\u679c\u3002<\/li>\n<\/ul>\n<p>\u8bc4\u4f30\u4efb\u52a1\u5b8c\u6210\u540e\uff0c\u7ed3\u679c\u4f1a\u9ed8\u8ba4\u4fdd\u5b58\u5728\u9879\u76ee\u76ee\u5f55\u4e0b\u7684\u00a0<code>.\/logs\/<\/code>\u00a0\u6587\u4ef6\u5939\u4e2d\u3002<\/p>\n<h3><strong>4. \u67e5\u770b\u8bc4\u4f30\u7ed3\u679c<\/strong><\/h3>\n<p>\u4f60\u6709\u4e24\u79cd\u65b9\u5f0f\u67e5\u770b\u7ed3\u679c\uff1a<\/p>\n<p><strong>\u65b9\u5f0f\u4e00\uff1a\u76f4\u63a5\u67e5\u770b\u65e5\u5fd7\u6587\u4ef6<\/strong><br \/>\n\u7ed3\u679c\u65e5\u5fd7\u662f\u7eaf\u6587\u672c\u6216JSON\u6587\u4ef6\uff0c\u4f60\u53ef\u4ee5\u76f4\u63a5\u7528\u6587\u672c\u7f16\u8f91\u5668\u6253\u5f00\u4f4d\u4e8e\u00a0<code>.\/logs\/<\/code>\u00a0\u76ee\u5f55\u4e0b\u7684\u65e5\u5fd7\u6587\u4ef6\u8fdb\u884c\u67e5\u770b\u3002<\/p>\n<p><strong>\u65b9\u5f0f\u4e8c\uff1a\u4f7f\u7528\u4ea4\u4e92\u5f0f\u754c\u9762<\/strong><br \/>\nOpenBench \u63d0\u4f9b\u4e86\u4e00\u4e2a\u66f4\u53cb\u597d\u7684\u4ea4\u4e92\u5f0f\u754c\u9762\u6765\u5c55\u793a\u7ed3\u679c\u3002<\/p>\n<pre><code>bench view\r\n```\u8be5\u547d\u4ee4\u4f1a\u542f\u52a8\u4e00\u4e2a\u672c\u5730\u670d\u52a1\uff0c\u8ba9\u4f60\u53ef\u4ee5\u5728\u6d4f\u89c8\u5668\u4e2d\u76f4\u89c2\u5730\u6d4f\u89c8\u548c\u5206\u6790\u5386\u6b21\u8bc4\u4f30\u7684\u7ed3\u679c\u3002\r\n### **5. \u4e3b\u8981\u547d\u4ee4\u548c\u5e38\u7528\u9009\u9879**\r\nOpenBench \u7684\u6838\u5fc3\u529f\u80fd\u901a\u8fc7 `bench` \u547d\u4ee4\u66b4\u9732\u3002\r\n- `bench --help`: \u663e\u793a\u6240\u6709\u53ef\u7528\u7684\u547d\u4ee4\u548c\u5168\u5c40\u9009\u9879\u3002\r\n- `bench list`: \u5217\u51fa\u6240\u6709\u53ef\u7528\u7684\u57fa\u51c6\u6d4b\u8bd5\u3001\u6a21\u578b\u548c\u6807\u5fd7\u3002\r\n- `bench eval &lt;benchmark&gt;`: \u8fd0\u884c\u6307\u5b9a\u7684\u57fa\u51c6\u6d4b\u8bd5\u3002\r\n- `bench view`: \u542f\u52a8\u4ea4\u4e92\u5f0f\u754c\u9762\u67e5\u770b\u65e5\u5fd7\u3002\r\n#### **`eval` \u547d\u4ee4\u7684\u5173\u952e\u9009\u9879**\r\n`eval` \u547d\u4ee4\u652f\u6301\u4e30\u5bcc\u7684\u9009\u9879\u6765\u63a7\u5236\u8bc4\u4f30\u6d41\u7a0b\uff0c\u4f60\u53ef\u4ee5\u901a\u8fc7\u547d\u4ee4\u884c\u53c2\u6570\u6216\u73af\u5883\u53d8\u91cf\u6765\u8bbe\u7f6e\u3002\r\n| \u9009\u9879 | \u73af\u5883\u53d8\u91cf | \u63cf\u8ff0 |\r\n| --- | --- | --- |\r\n| `--model` | `BENCH_MODEL` | \u6307\u5b9a\u8981\u8bc4\u4f30\u7684\u4e00\u4e2a\u6216\u591a\u4e2a\u6a21\u578b\u3002 |\r\n| `--limit` | `BENCH_LIMIT` | \u9650\u5236\u8bc4\u4f30\u6837\u672c\u7684\u6570\u91cf\uff0c\u53ef\u4ee5\u662f\u5177\u4f53\u6570\u5b57\u6216\u8303\u56f4\uff08\u5982`10,20`\uff09\u3002 |\r\n| `--temperature` | `BENCH_TEMPERATURE` | \u8bbe\u7f6e\u6a21\u578b\u7684\u751f\u6210\u6e29\u5ea6\uff0c\u5f71\u54cd\u8f93\u51fa\u7684\u968f\u673a\u6027\u3002 |\r\n| `--max-connections`| `BENCH_MAX_CONNECTIONS`| \u8bbe\u7f6e\u4e0e\u6a21\u578bAPI\u7684\u6700\u5927\u5e76\u884c\u8fde\u63a5\u6570\uff0c\u9ed8\u8ba4\u4e3a10\u3002 |\r\n| `--logfile` | `BENCH_OUTPUT` | \u6307\u5b9a\u4fdd\u5b58\u7ed3\u679c\u7684\u65e5\u5fd7\u6587\u4ef6\u8def\u5f84\u3002 |\r\n| `--sandbox` | `BENCH_SANDBOX` | \u6307\u5b9a\u4ee3\u7801\u6267\u884c\u7684\u73af\u5883\uff0c\u5982`local`\u6216`docker`\uff0c\u7528\u4e8eHumanEval\u7b49\u7f16\u7801\u6d4b\u8bd5\u3002 |\r\n| `--json` | \u65e0 | \u5982\u679c\u8bbe\u7f6e\u6b64\u9879\uff0c\u7ed3\u679c\u5c06\u4ee5JSON\u683c\u5f0f\u8f93\u51fa\u3002 |\r\n### **6. \u4f7f\u7528\u4e0d\u540c\u4f9b\u5e94\u5546\u6216\u672c\u5730\u6a21\u578b**\r\n\u4f60\u53ef\u4ee5\u8f7b\u677e\u5207\u6362\u4e0d\u540c\u7684\u6a21\u578b\u4f9b\u5e94\u5546\u3002\r\n```bash\r\n# \u4f7f\u7528 OpenAI \u7684\u6a21\u578b\r\nbench eval humaneval --model openai\/o3-2025-04-16\r\n# \u4f7f\u7528 Google \u7684\u6a21\u578b\r\nbench eval mmlu --model google\/gemini-2.5-pro\r\n# \u4f7f\u7528\u901a\u8fc7 <a href=\"https:\/\/www.kdjingpai.com\/ollama\/\">Ollama<\/a> \u8fd0\u884c\u7684\u672c\u5730\u6a21\u578b\r\n# \u786e\u4fdd Ollama \u670d\u52a1\u6b63\u5728\u8fd0\u884c\r\nbench eval musr --model ollama\/llama3.1:70b\r\n<\/code><\/pre>\n<h3><strong>7. \u5904\u7406 Hugging Face \u6570\u636e\u96c6\u4e0b\u8f7d\u95ee\u9898<\/strong><\/h3>\n<p>\u90e8\u5206\u57fa\u51c6\u6d4b\u8bd5\u9700\u8981\u4ece Hugging Face \u4e0b\u8f7d\u6570\u636e\u96c6\u3002\u5982\u679c\u9047\u5230\u201cgated\u201d\u9519\u8bef\uff0c\u8bf4\u660e\u8be5\u6570\u636e\u96c6\u9700\u8981\u7528\u6237\u8ba4\u8bc1\u3002\u4f60\u9700\u8981\u8bbe\u7f6e Hugging Face \u7684\u8bbf\u95ee\u4ee4\u724c\u3002<\/p>\n<pre><code>export HF_TOKEN=\"\u4f60\u7684HuggingFace\u4ee4\u724c\"\r\n<\/code><\/pre>\n<p>\u5b8c\u6210\u4ee5\u4e0a\u6b65\u9aa4\u540e\uff0c\u91cd\u65b0\u8fd0\u884c\u00a0<code>bench eval<\/code>\u00a0\u547d\u4ee4\u5373\u53ef\u89e3\u51b3\u95ee\u9898\u3002<\/p>\n<h2>\u5e94\u7528\u573a\u666f<\/h2>\n<ol>\n<li><strong>\u6a21\u578b\u7814\u7a76\u4e0e\u5f00\u53d1<\/strong><br \/>\n\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u5728\u5f00\u53d1\u65b0\u7684\u8bed\u8a00\u6a21\u578b\u65f6\uff0c\u53ef\u4ee5\u4f7f\u7528OpenBench\u5feb\u901f\u5728\u591a\u4e2a\u884c\u4e1a\u6807\u51c6\u57fa\u51c6\u4e0a\u6d4b\u8bd5\u65b0\u6a21\u578b\u7684\u6027\u80fd\uff0c\u5e76\u4e0e\u73b0\u6709\u4e3b\u6d41\u6a21\u578b\u8fdb\u884c\u91cf\u5316\u6bd4\u8f83\uff0c\u4ece\u800c\u9a8c\u8bc1\u6a21\u578b\u7684\u6539\u8fdb\u6548\u679c\u3002<\/li>\n<li><strong>\u6a21\u578b\u9009\u578b\u4e0e\u91c7\u8d2d<\/strong><br \/>\n\u4f01\u4e1a\u6216\u56e2\u961f\u5728\u9009\u62e9\u9002\u5408\u81ea\u8eab\u4e1a\u52a1\u7684\u8bed\u8a00\u6a21\u578b\u65f6\uff0c\u53ef\u4ee5\u5229\u7528OpenBench\u5bf9\u6765\u81ea\u4e0d\u540c\u4f9b\u5e94\u5546\uff08\u5982OpenAI, Google, Anthropic\uff09\u7684\u5019\u9009\u6a21\u578b\u8fdb\u884c\u7edf\u4e00\u3001\u516c\u5e73\u7684\u6027\u80fd\u8bc4\u4f30\uff0c\u4ee5\u6570\u636e\u9a71\u52a8\u51b3\u7b56\u3002<\/li>\n<li><strong>\u6301\u7eed\u96c6\u6210\u4e0e\u56de\u5f52\u6d4b\u8bd5<\/strong><br \/>\n\u5bf9\u4e8e\u9700\u8981\u9891\u7e41\u5fae\u8c03\u6216\u8fed\u4ee3\u6a21\u578b\u7684\u573a\u666f\uff0c\u53ef\u4ee5\u5c06OpenBench\u96c6\u6210\u5230CI\/CD\u6d41\u7a0b\u4e2d\u3002\u6bcf\u5f53\u6a21\u578b\u66f4\u65b0\u540e\uff0c\u81ea\u52a8\u8fd0\u884c\u4e00\u5957\u6807\u51c6\u5316\u7684\u57fa\u51c6\u6d4b\u8bd5\uff0c\u4ee5\u786e\u4fdd\u6a21\u578b\u6027\u80fd\u6ca1\u6709\u51fa\u73b0\u610f\u5916\u8870\u9000\u3002<\/li>\n<li><strong>\u672c\u5730\u6a21\u578b\u6027\u80fd\u8bc4\u4f30<\/strong><br \/>\n\u5bf9\u4e8e\u6ce8\u91cd\u6570\u636e\u9690\u79c1\u6216\u9700\u8981\u79bb\u7ebf\u8fd0\u884c\u7684\u573a\u666f\uff0c\u5f00\u53d1\u8005\u53ef\u4ee5\u4f7f\u7528Ollama\u5728\u672c\u5730\u90e8\u7f72\u5f00\u6e90\u6a21\u578b\u3002OpenBench\u53ef\u4ee5\u8fde\u63a5\u5230\u672c\u5730Ollama\u670d\u52a1\uff0c\u5bf9\u8fd9\u4e9b\u672c\u5730\u6a21\u578b\u7684\u77e5\u8bc6\u3001\u63a8\u7406\u548c\u7f16\u7801\u80fd\u529b\u8fdb\u884c\u5168\u9762\u8bc4\u4f30\u3002<\/li>\n<\/ol>\n<h2>QA<\/h2>\n<ol>\n<li><strong>OpenBench \u4e0e Inspect AI \u6709\u4ec0\u4e48\u533a\u522b\uff1f<\/strong><br \/>\nOpenBench \u662f\u4e00\u4e2a\u6784\u5efa\u5728 Inspect AI \u6846\u67b6\u4e4b\u4e0a\u7684\u57fa\u51c6\u6d4b\u8bd5\u5e93\u3002\u53ef\u4ee5\u7406\u89e3\u4e3a\uff0cInspect AI \u63d0\u4f9b\u4e86\u5e95\u5c42\u7684\u8bc4\u4f30\u80fd\u529b\u548c\u5de5\u5177\uff0c\u800c OpenBench \u5728\u6b64\u57fa\u7840\u4e0a\u63d0\u4f9b\u4e8620\u591a\u4e2a\u4e3b\u6d41\u57fa\u51c6\u6d4b\u8bd5\u7684\u73b0\u6210\u5b9e\u73b0\u3001\u7edf\u4e00\u7684\u547d\u4ee4\u884c\u5de5\u5177\u4ee5\u53ca\u8de8\u8bc4\u6d4b\u5171\u4eab\u7684\u5b9e\u7528\u7a0b\u5e8f\uff08\u5982\u6570\u5b66\u8bc4\u5206\u5668\uff09\u3002OpenBench \u4e13\u6ce8\u4e8e\u7b80\u5316\u8fd0\u884c\u6807\u51c6\u5316\u57fa\u51c6\u6d4b\u8bd5\u7684\u6d41\u7a0b\u548c\u63d0\u5347\u5f00\u53d1\u8005\u7684\u4f7f\u7528\u4f53\u9a8c\u3002<\/li>\n<li><strong>\u4e3a\u4ec0\u4e48\u8981\u9009\u62e9 OpenBench\uff0c\u800c\u4e0d\u662f lm-evaluation-harness \u6216 lighteval \u7b49\u5176\u4ed6\u5de5\u5177\uff1f<\/strong><br \/>\n\u8fd9\u51e0\u4e2a\u5de5\u5177\u5404\u6709\u4fa7\u91cd\u3002OpenBench \u7684\u4e3b\u8981\u4f18\u52bf\u5728\u4e8e\u5176\u6e05\u6670\u3001\u6613\u4e8e\u7406\u89e3\u548c\u4fee\u6539\u7684\u8bc4\u6d4b\u5b9e\u73b0\u3002\u5b83\u901a\u8fc7\u5171\u4eab\u7ec4\u4ef6\u51cf\u5c11\u4e86\u4e0d\u540c\u57fa\u51c6\u6d4b\u8bd5\u95f4\u7684\u4ee3\u7801\u91cd\u590d\uff0c\u5e76\u901a\u8fc7\u7b80\u6d01\u7684\u547d\u4ee4\u884c\u5de5\u5177\u548c\u4e00\u81f4\u7684\u8bbe\u8ba1\u6a21\u5f0f\uff0c\u4f18\u5316\u4e86\u5f00\u53d1\u8005\u7684\u4f7f\u7528\u4f53\u9a8c\u3002\u5982\u679c\u4f60\u9700\u8981\u4e00\u4e2a\u6613\u4e8e\u6269\u5c55\u548c\u7ef4\u62a4\u3001\u8bc4\u6d4b\u4ee3\u7801\u53ef\u8bfb\u6027\u9ad8\u7684\u5de5\u5177\uff0cOpenBench \u662f\u4e00\u4e2a\u5f88\u597d\u7684\u9009\u62e9\u3002<\/li>\n<li><strong>\u5982\u4f55\u5728\u865a\u62df\u73af\u5883\u4e4b\u5916\u4f7f\u7528\u00a0<code>bench<\/code>\u00a0\u547d\u4ee4\uff1f<\/strong><br \/>\n\u5982\u679c\u4f60\u5e0c\u671b\u5728\u7cfb\u7edf\u4e2d\u7684\u4efb\u4f55\u8def\u5f84\u4e0b\u90fd\u80fd\u76f4\u63a5\u8c03\u7528\u00a0<code>bench<\/code>\u00a0\u547d\u4ee4\uff0c\u800c\u4e0d\u662f\u6bcf\u6b21\u90fd\u8981\u6fc0\u6d3b\u865a\u62df\u73af\u5883\uff0c\u53ef\u4ee5\u5728\u9879\u76ee\u514b\u9686\u5230\u672c\u5730\u540e\uff0c\u8fd0\u884c\u4ee5\u4e0b\u547d\u4ee4\u8fdb\u884c\u53ef\u7f16\u8f91\u6a21\u5f0f\u7684\u5b89\u88c5\uff1a\u00a0<code>uv run pip install -e .<\/code>\u3002<\/li>\n<li><strong>\u8fd0\u884c\u8bc4\u6d4b\u65f6\u63d0\u793a HuggingFace \u9700\u8981\u767b\u5f55\uff0c\u5982\u4f55\u89e3\u51b3\uff1f<\/strong><br \/>\n\u8fd9\u901a\u5e38\u662f\u56e0\u4e3a\u8bc4\u6d4b\u6240\u9700\u7684\u6570\u636e\u96c6\u5728 HuggingFace \u4e0a\u662f\u53d7\u4fdd\u62a4\u7684\uff08gated\uff09\u3002\u4f60\u9700\u8981\u5148\u83b7\u53d6\u4e00\u4e2a HuggingFace \u7684\u8bbf\u95ee\u4ee4\u724c\uff08token\uff09\uff0c\u7136\u540e\u901a\u8fc7\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf\u00a0<code>HF_TOKEN<\/code>\u00a0\u6765\u89e3\u51b3\u3002\u4f8b\u5982\uff1a<code>export HF_TOKEN=\"hf_xxxxxxxx\"<\/code>\uff0c\u4e4b\u540e\u518d\u91cd\u65b0\u8fd0\u884c\u8bc4\u6d4b\u547d\u4ee4\u5373\u53ef\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>OpenBench\u662f\u4e00\u4e2a\u5f00\u6e90\u7684\u8bed\u8a00\u6a21\u578b\u8bc4\u4f30\u5de5\u5177\uff0c\u5b83\u4e0d\u53d7\u9650\u4e8e\u7279\u5b9a\u7684\u6a21\u578b\u4f9b\u5e94\u5546\u3002\u5f00\u53d1\u8005\u53ef\u4ee5\u4f7f\u7528\u5b83\u5728\u8d85\u8fc720\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u96c6\u4e0a\u5bf9\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u6807\u51c6\u5316\u7684\u3001\u53ef\u590d\u73b0\u7684\u6027\u80fd\u8bc4\u6d4b\uff0c\u8fd9\u4e9b\u6d4b\u8bd5\u6db5\u76d6\u4e86\u77e5\u8bc6\u3001\u63a8\u7406\u3001\u7f16\u7801\u548c\u6570\u5b66\u7b49\u591a\u4e2a\u9886\u57df\u3002OpenBench\u7684\u6838\u5fc3\u4f18\u52bf\u5728\u4e8e\u5176&#8230;<\/p>\n","protected":false},"author":1,"featured_media":32782,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[425,20],"tags":[230],"class_list":["post-34618","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-professional","category-tool","tag-aikaiyuanxiangmu"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/34618","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/comments?post=34618"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/34618\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media\/32782"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=34618"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=34618"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=34618"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}