{"id":29641,"date":"2025-03-30T18:00:04","date_gmt":"2025-03-30T10:00:04","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=29641"},"modified":"2025-03-30T18:00:04","modified_gmt":"2025-03-30T10:00:04","slug":"jiemidamoxingbang","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/jiemidamoxingbang\/","title":{"rendered":"\u63ed\u79d8\u5927\u6a21\u578b\u201c\u5e7b\u89c9\u201d\uff1aHHEM \u6392\u884c\u699c\u900f\u89c6 LLM \u4e8b\u5b9e\u4e00\u81f4\u6027\u73b0\u72b6"},"content":{"rendered":"<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u80fd\u529b\u65e5\u65b0\u6708\u5f02\uff0c\u4f46\u5176\u8f93\u51fa\u5185\u5bb9\u4e2d\u51fa\u73b0\u4e8b\u5b9e\u6027\u9519\u8bef\u6216\u4e0e\u539f\u6587\u65e0\u5173\u4fe1\u606f\u7684\u201c\u5e7b\u89c9\u201d\u73b0\u8c61\uff0c\u59cb\u7ec8\u662f\u963b\u788d\u5176\u66f4\u5e7f\u6cdb\u5e94\u7528\u548c\u6df1\u5ea6\u4fe1\u4efb\u7684\u4e00\u5927\u96be\u9898\u3002\u4e3a\u4e86\u91cf\u5316\u8bc4\u4f30\u8fd9\u4e00\u95ee\u9898\uff0c<a href=\"https:\/\/huggingface.co\/spaces\/vectara\/leaderboard\">Hughes Hallucination Evaluation Model (HHEM) \u6392\u884c\u699c<\/a>\u5e94\u8fd0\u800c\u751f\uff0c\u4e13\u6ce8\u4e8e\u8861\u91cf\u4e3b\u6d41 LLM \u5728\u751f\u6210\u6587\u6863\u6458\u8981\u65f6\u7684\u5e7b\u89c9\u9891\u7387\u3002<\/p>\n<p>\u6240\u8c13\u201c\u5e7b\u89c9\u201d\uff0c\u6307\u7684\u662f\u6a21\u578b\u5728\u6458\u8981\u4e2d\u5f15\u5165\u4e86\u539f\u59cb\u6587\u6863\u5e76\u672a\u5305\u542b\u3001\u751a\u81f3\u76f8\u6096\u7684\u201c\u4e8b\u5b9e\u201d\u3002\u8fd9\u5bf9\u4e8e\u4f9d\u8d56 LLM \u8fdb\u884c\u4fe1\u606f\u5904\u7406\uff0c\u5c24\u5176\u662f\u57fa\u4e8e\u68c0\u7d22\u589e\u5f3a\u751f\u6210\uff08RAG\uff09\u7684\u5e94\u7528\u573a\u666f\u6765\u8bf4\uff0c\u662f\u4e00\u4e2a\u5173\u952e\u7684\u8d28\u91cf\u74f6\u9888\u3002\u6bd5\u7adf\uff0c\u5982\u679c\u6a21\u578b\u4e0d\u80fd\u5fe0\u5b9e\u4e8e\u7ed9\u5b9a\u4fe1\u606f\uff0c\u90a3\u4e48\u5176\u8f93\u51fa\u7684\u53ef\u4fe1\u5ea6\u5c31\u5927\u6253\u6298\u6263\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>HHEM \u5982\u4f55\u5de5\u4f5c\uff1f<\/h3>\n<p>\u8be5\u6392\u884c\u699c\u91c7\u7528 Vectara \u516c\u53f8\u5f00\u53d1\u7684 HHEM-2.1 \u5e7b\u89c9\u8bc4\u4f30\u6a21\u578b\u3002\u5176\u5de5\u4f5c\u539f\u7406\u662f\uff0c\u9488\u5bf9\u4e00\u4efd\u6e90\u6587\u6863\u548c\u7531\u7279\u5b9a LLM \u751f\u6210\u7684\u6458\u8981\uff0cHHEM \u6a21\u578b\u4f1a\u8f93\u51fa\u4e00\u4e2a\u4ecb\u4e8e 0 \u5230 1 \u4e4b\u95f4\u7684\u5e7b\u89c9\u5206\u6570\u3002\u5206\u6570\u8d8a\u63a5\u8fd1 1\uff0c\u8868\u793a\u6458\u8981\u4e0e\u6e90\u6587\u6863\u7684\u4e8b\u5b9e\u4e00\u81f4\u6027\u8d8a\u9ad8\uff1b\u8d8a\u63a5\u8fd1 0\uff0c\u5219\u8868\u793a\u5e7b\u89c9\u8d8a\u4e25\u91cd\uff0c\u751a\u81f3\u5b8c\u5168\u662f\u7f16\u9020\u7684\u5185\u5bb9\u3002Vectara \u4e5f\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5f00\u6e90\u7248\u672c HHEM-2.1-Open\uff0c\u53ef\u4f9b\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u5728\u672c\u5730\u8fdb\u884c\u8bc4\u4f30\uff0c\u5176\u6a21\u578b\u5361\u53d1\u5e03\u5728 Hugging Face \u5e73\u53f0\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u8bc4\u4f30\u57fa\u51c6<\/h3>\n<p>\u8bc4\u4f30\u4f7f\u7528\u4e86\u4e00\u4e2a\u5305\u542b 1006 \u4efd\u6587\u6863\u7684\u6570\u636e\u96c6\uff0c\u8fd9\u4e9b\u6587\u6863\u4e3b\u8981\u6765\u6e90\u4e8e\u516c\u5f00\u6570\u636e\u96c6\uff0c\u5982\u7ecf\u5178\u7684 CNN\/Daily Mail Corpus\u3002\u9879\u76ee\u56e2\u961f\u4f7f\u7528\u53c2\u4e0e\u8bc4\u4f30\u7684\u5404\u4e2a LLM \u4e3a\u6bcf\u4efd\u6587\u6863\u751f\u6210\u6458\u8981\uff0c\u7136\u540e\u8ba1\u7b97\u6bcf\u5bf9\uff08\u6e90\u6587\u6863\uff0c\u751f\u6210\u6458\u8981\uff09\u7684 HHEM \u5206\u6570\u3002\u4e3a\u4e86\u4fdd\u8bc1\u8bc4\u4f30\u7684\u6807\u51c6\u5316\uff0c\u6240\u6709\u6a21\u578b\u8c03\u7528\u5747\u8bbe\u7f6e\u00a0<code>temperature<\/code>\u00a0\u53c2\u6570\u4e3a 0\uff0c\u65e8\u5728\u83b7\u53d6\u6a21\u578b\u6700\u5177\u786e\u5b9a\u6027\u7684\u8f93\u51fa\u3002<\/p>\n<p>\u8bc4\u4f30\u6307\u6807\u4e3b\u8981\u5305\u62ec\uff1a<\/p>\n<ul>\n<li><strong>\u5e7b\u89c9\u7387 (Hallucination Rate):<\/strong>\u00a0HHEM \u5206\u6570\u4f4e\u4e8e 0.5 \u7684\u6458\u8981\u6240\u5360\u7684\u767e\u5206\u6bd4\u3002\u8fd9\u4e2a\u503c\u8d8a\u4f4e\u8d8a\u597d\u3002<\/li>\n<li><strong>\u4e8b\u5b9e\u4e00\u81f4\u6027\u7387 (Factual Consistency Rate):<\/strong>\u00a0100% \u51cf\u53bb\u5e7b\u89c9\u7387\uff0c\u53cd\u6620\u4e86\u6458\u8981\u5185\u5bb9\u5fe0\u5b9e\u4e8e\u539f\u6587\u7684\u6bd4\u4f8b\u3002<\/li>\n<li><strong>\u56de\u7b54\u7387 (Answer Rate):<\/strong>\u00a0\u6a21\u578b\u6210\u529f\u751f\u6210\u975e\u7a7a\u6458\u8981\u7684\u767e\u5206\u6bd4\u3002\u90e8\u5206\u6a21\u578b\u53ef\u80fd\u56e0\u5185\u5bb9\u5b89\u5168\u7b56\u7565\u6216\u5176\u4ed6\u539f\u56e0\u62d2\u7edd\u56de\u7b54\u6216\u51fa\u9519\u3002<\/li>\n<li><strong>\u5e73\u5747\u6458\u8981\u957f\u5ea6 (Average Summary Length):<\/strong>\u00a0\u751f\u6210\u6458\u8981\u7684\u5e73\u5747\u8bcd\u6570\uff0c\u53ef\u4fa7\u9762\u53cd\u6620\u6a21\u578b\u7684\u8f93\u51fa\u98ce\u683c\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h3>LLM \u5e7b\u89c9\u6392\u884c\u699c\u89e3\u8bfb<\/h3>\n<p>\u4ee5\u4e0b\u662f\u57fa\u4e8e HHEM-2.1 \u6a21\u578b\u8bc4\u4f30\u5f97\u51fa\u7684 LLM \u5e7b\u89c9\u6392\u884c\u699c\uff08\u6570\u636e\u622a\u81f3 2025 \u5e74 3 \u6708 25 \u65e5\uff0c\u8bf7\u4ee5\u5b9e\u9645\u66f4\u65b0\u4e3a\u51c6\uff09\uff1a<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/03\/841183cde8fc0f1.png\" alt=\"\" \/><\/p>\n<p>&nbsp;<\/p>\n<table>\n<thead>\n<tr>\n<th>Model<\/th>\n<th align=\"right\">Hallucination Rate<\/th>\n<th align=\"right\">Factual Consistency Rate<\/th>\n<th align=\"right\">Answer Rate<\/th>\n<th align=\"right\">Average Summary Length (Words)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Google Gemini-2.0-Flash-001<\/td>\n<td align=\"right\">0.7 %<\/td>\n<td align=\"right\">99.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">65.2<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-2.0-Pro-Exp<\/td>\n<td align=\"right\">0.8 %<\/td>\n<td align=\"right\">99.2 %<\/td>\n<td align=\"right\">99.7 %<\/td>\n<td align=\"right\">61.5<\/td>\n<\/tr>\n<tr>\n<td>OpenAI-o3-mini-high-reasoning<\/td>\n<td align=\"right\">0.8 %<\/td>\n<td align=\"right\">99.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">79.5<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-2.5-Pro-Exp-0325<\/td>\n<td align=\"right\">1.1 %<\/td>\n<td align=\"right\">98.9 %<\/td>\n<td align=\"right\">95.1 %<\/td>\n<td align=\"right\">72.9<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-2.0-Flash-Lite-Preview<\/td>\n<td align=\"right\">1.2 %<\/td>\n<td align=\"right\">98.8 %<\/td>\n<td align=\"right\">99.5 %<\/td>\n<td align=\"right\">60.9<\/td>\n<\/tr>\n<tr>\n<td>OpenAI-GPT-4.5-Preview<\/td>\n<td align=\"right\">1.2 %<\/td>\n<td align=\"right\">98.8 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">77.0<\/td>\n<\/tr>\n<tr>\n<td>Zhipu AI GLM-4-9B-Chat<\/td>\n<td align=\"right\">1.3 %<\/td>\n<td align=\"right\">98.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">58.1<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-2.0-Flash-Exp<\/td>\n<td align=\"right\">1.3 %<\/td>\n<td align=\"right\">98.7 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">60.0<\/td>\n<\/tr>\n<tr>\n<td>OpenAI-o1-mini<\/td>\n<td align=\"right\">1.4 %<\/td>\n<td align=\"right\">98.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">78.3<\/td>\n<\/tr>\n<tr>\n<td>GPT-4o<\/td>\n<td align=\"right\">1.5 %<\/td>\n<td align=\"right\">98.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">77.8<\/td>\n<\/tr>\n<tr>\n<td>Amazon Nova-Micro-V1<\/td>\n<td align=\"right\">1.6 %<\/td>\n<td align=\"right\">98.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">90.0<\/td>\n<\/tr>\n<tr>\n<td>GPT-4o-mini<\/td>\n<td align=\"right\">1.7 %<\/td>\n<td align=\"right\">98.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">76.3<\/td>\n<\/tr>\n<tr>\n<td>GPT-4-Turbo<\/td>\n<td align=\"right\">1.7 %<\/td>\n<td align=\"right\">98.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">86.2<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-2.0-Flash-Thinking-Exp<\/td>\n<td align=\"right\">1.8 %<\/td>\n<td align=\"right\">98.2 %<\/td>\n<td align=\"right\">99.3 %<\/td>\n<td align=\"right\">73.2<\/td>\n<\/tr>\n<tr>\n<td>Amazon Nova-Lite-V1<\/td>\n<td align=\"right\">1.8 %<\/td>\n<td align=\"right\">98.2 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">80.7<\/td>\n<\/tr>\n<tr>\n<td>GPT-4<\/td>\n<td align=\"right\">1.8 %<\/td>\n<td align=\"right\">98.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">81.1<\/td>\n<\/tr>\n<tr>\n<td>Amazon Nova-Pro-V1<\/td>\n<td align=\"right\">1.8 %<\/td>\n<td align=\"right\">98.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">85.5<\/td>\n<\/tr>\n<tr>\n<td>GPT-3.5-Turbo<\/td>\n<td align=\"right\">1.9 %<\/td>\n<td align=\"right\">98.1 %<\/td>\n<td align=\"right\">99.6 %<\/td>\n<td align=\"right\">84.1<\/td>\n<\/tr>\n<tr>\n<td>XAI-2<\/td>\n<td align=\"right\">1.9 %<\/td>\n<td align=\"right\">98.1<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">86.5<\/td>\n<\/tr>\n<tr>\n<td>AI21 Jamba-1.6-Large<\/td>\n<td align=\"right\">2.3 %<\/td>\n<td align=\"right\">97.7 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">85.6<\/td>\n<\/tr>\n<tr>\n<td>OpenAI O1-Pro<\/td>\n<td align=\"right\">2.4 %<\/td>\n<td align=\"right\">97.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">81.0<\/td>\n<\/tr>\n<tr>\n<td>OpenAI-o1<\/td>\n<td align=\"right\">2.4 %<\/td>\n<td align=\"right\">97.6 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">73.0<\/td>\n<\/tr>\n<tr>\n<td>DeepSeek-V2.5<\/td>\n<td align=\"right\">2.4 %<\/td>\n<td align=\"right\">97.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">83.2<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Orca-2-13b<\/td>\n<td align=\"right\">2.5 %<\/td>\n<td align=\"right\">97.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">66.2<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-3.5-MoE-instruct<\/td>\n<td align=\"right\">2.5 %<\/td>\n<td align=\"right\">97.5 %<\/td>\n<td align=\"right\">96.3 %<\/td>\n<td align=\"right\">69.7<\/td>\n<\/tr>\n<tr>\n<td>Intel Neural-Chat-7B-v3-3<\/td>\n<td align=\"right\">2.6 %<\/td>\n<td align=\"right\">97.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">60.7<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-3-12B-Instruct<\/td>\n<td align=\"right\">2.8 %<\/td>\n<td align=\"right\">97.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">69.6<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-7B-Instruct<\/td>\n<td align=\"right\">2.8 %<\/td>\n<td align=\"right\">97.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">71.0<\/td>\n<\/tr>\n<tr>\n<td>AI21 Jamba-1.5-Mini<\/td>\n<td align=\"right\">2.9 %<\/td>\n<td align=\"right\">97.1 %<\/td>\n<td align=\"right\">95.6 %<\/td>\n<td align=\"right\">74.5<\/td>\n<\/tr>\n<tr>\n<td>XAI-2-Vision<\/td>\n<td align=\"right\">2.9 %<\/td>\n<td align=\"right\">97.1<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">79.8<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-Max<\/td>\n<td align=\"right\">2.9 %<\/td>\n<td align=\"right\">97.1 %<\/td>\n<td align=\"right\">88.8 %<\/td>\n<td align=\"right\">90.4<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-3-27B-Instruct<\/td>\n<td align=\"right\">3.0 %<\/td>\n<td align=\"right\">97.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">62.5<\/td>\n<\/tr>\n<tr>\n<td>Snowflake-Arctic-Instruct<\/td>\n<td align=\"right\">3.0 %<\/td>\n<td align=\"right\">97.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">68.7<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-32B-Instruct<\/td>\n<td align=\"right\">3.0 %<\/td>\n<td align=\"right\">97.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">67.9<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-3-mini-128k-instruct<\/td>\n<td align=\"right\">3.1 %<\/td>\n<td align=\"right\">96.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">60.1<\/td>\n<\/tr>\n<tr>\n<td><a href=\"https:\/\/www.kdjingpai.com\/le-chat-mistral\/\">Mistral<\/a> Small3<\/td>\n<td align=\"right\">3.1 %<\/td>\n<td align=\"right\">96.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">74.9<\/td>\n<\/tr>\n<tr>\n<td>OpenAI-o1-preview<\/td>\n<td align=\"right\">3.3 %<\/td>\n<td align=\"right\">96.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">119.3<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-1.5-Flash-002<\/td>\n<td align=\"right\">3.4 %<\/td>\n<td align=\"right\">96.6 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">59.4<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-4-mini-instruct<\/td>\n<td align=\"right\">3.4 %<\/td>\n<td align=\"right\">96.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">69.7<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-3-4B-Instruct<\/td>\n<td align=\"right\">3.7 %<\/td>\n<td align=\"right\">96.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">63.7<\/td>\n<\/tr>\n<tr>\n<td>01-AI Yi-1.5-34B-Chat<\/td>\n<td align=\"right\">3.7 %<\/td>\n<td align=\"right\">96.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">83.7<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.1-405B-Instruct<\/td>\n<td align=\"right\">3.9 %<\/td>\n<td align=\"right\">96.1 %<\/td>\n<td align=\"right\">99.6 %<\/td>\n<td align=\"right\">85.7<\/td>\n<\/tr>\n<tr>\n<td>DeepSeek-V3<\/td>\n<td align=\"right\">3.9 %<\/td>\n<td align=\"right\">96.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">88.2<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-3-mini-4k-instruct<\/td>\n<td align=\"right\">4.0 %<\/td>\n<td align=\"right\">96.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">86.8<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.3-70B-Instruct<\/td>\n<td align=\"right\">4.0 %<\/td>\n<td align=\"right\">96.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">85.3<\/td>\n<\/tr>\n<tr>\n<td>InternLM3-8B-Instruct<\/td>\n<td align=\"right\">4.0 %<\/td>\n<td align=\"right\">96.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">97.5<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-3.5-mini-instruct<\/td>\n<td align=\"right\">4.1 %<\/td>\n<td align=\"right\">95.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">75.0<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Large2<\/td>\n<td align=\"right\">4.1 %<\/td>\n<td align=\"right\">95.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">77.4<\/td>\n<\/tr>\n<tr>\n<td>Llama-3-70B-Chat-hf<\/td>\n<td align=\"right\">4.1 %<\/td>\n<td align=\"right\">95.9 %<\/td>\n<td align=\"right\">99.2 %<\/td>\n<td align=\"right\">68.5<\/td>\n<\/tr>\n<tr>\n<td>Qwen2-VL-7B-Instruct<\/td>\n<td align=\"right\">4.2 %<\/td>\n<td align=\"right\">95.8 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">73.9<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-14B-Instruct<\/td>\n<td align=\"right\">4.2 %<\/td>\n<td align=\"right\">95.8 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">74.8<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-72B-Instruct<\/td>\n<td align=\"right\">4.3 %<\/td>\n<td align=\"right\">95.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">80.0<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.2-90B-Vision-Instruct<\/td>\n<td align=\"right\">4.3 %<\/td>\n<td align=\"right\">95.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">79.8<\/td>\n<\/tr>\n<tr>\n<td>Claude-3.7-Sonnet<\/td>\n<td align=\"right\">4.4 %<\/td>\n<td align=\"right\">95.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">97.8<\/td>\n<\/tr>\n<tr>\n<td>Claude-3.7-Sonnet-Think<\/td>\n<td align=\"right\">4.5 %<\/td>\n<td align=\"right\">95.5 %<\/td>\n<td align=\"right\">99.8 %<\/td>\n<td align=\"right\">99.9<\/td>\n<\/tr>\n<tr>\n<td><a href=\"https:\/\/www.kdjingpai.com\/cohere\/\">Cohere<\/a> Command-A<\/td>\n<td align=\"right\">4.5 %<\/td>\n<td align=\"right\">95.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">77.3<\/td>\n<\/tr>\n<tr>\n<td>AI21 Jamba-1.6-Mini<\/td>\n<td align=\"right\">4.6 %<\/td>\n<td align=\"right\">95.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">82.3<\/td>\n<\/tr>\n<tr>\n<td>XAI <a href=\"https:\/\/www.kdjingpai.com\/grok\/\">Grok<\/a><\/td>\n<td align=\"right\">4.6 %<\/td>\n<td align=\"right\">95.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">91.0<\/td>\n<\/tr>\n<tr>\n<td><a href=\"https:\/\/www.kdjingpai.com\/claudeanquanfubai\/\">Anthropic<\/a> Claude-3-5-sonnet<\/td>\n<td align=\"right\">4.6 %<\/td>\n<td align=\"right\">95.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">95.9<\/td>\n<\/tr>\n<tr>\n<td>Qwen2-72B-Instruct<\/td>\n<td align=\"right\">4.7 %<\/td>\n<td align=\"right\">95.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">100.1<\/td>\n<\/tr>\n<tr>\n<td>Microsoft Phi-4<\/td>\n<td align=\"right\">4.7 %<\/td>\n<td align=\"right\">95.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">100.3<\/td>\n<\/tr>\n<tr>\n<td>Mixtral-8x22B-Instruct-v0.1<\/td>\n<td align=\"right\">4.7 %<\/td>\n<td align=\"right\">95.3 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">92.0<\/td>\n<\/tr>\n<tr>\n<td>Anthropic Claude-3-5-haiku<\/td>\n<td align=\"right\">4.9 %<\/td>\n<td align=\"right\">95.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">92.9<\/td>\n<\/tr>\n<tr>\n<td>01-AI Yi-1.5-9B-Chat<\/td>\n<td align=\"right\">4.9 %<\/td>\n<td align=\"right\">95.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">85.7<\/td>\n<\/tr>\n<tr>\n<td>Cohere Command-R<\/td>\n<td align=\"right\">4.9 %<\/td>\n<td align=\"right\">95.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">68.7<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.1-70B-Instruct<\/td>\n<td align=\"right\">5.0 %<\/td>\n<td align=\"right\">95.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">79.6<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-3-1B-Instruct<\/td>\n<td align=\"right\">5.3 %<\/td>\n<td align=\"right\">94.7 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">57.9<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.1-8B-Instruct<\/td>\n<td align=\"right\">5.4 %<\/td>\n<td align=\"right\">94.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">71.0<\/td>\n<\/tr>\n<tr>\n<td>Cohere Command-R-Plus<\/td>\n<td align=\"right\">5.4 %<\/td>\n<td align=\"right\">94.6 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">68.4<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Small-3.1-24B-Instruct<\/td>\n<td align=\"right\">5.6 %<\/td>\n<td align=\"right\">94.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">73.1<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.2-11B-Vision-Instruct<\/td>\n<td align=\"right\">5.5 %<\/td>\n<td align=\"right\">94.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">67.3<\/td>\n<\/tr>\n<tr>\n<td>Llama-2-70B-Chat-hf<\/td>\n<td align=\"right\">5.9 %<\/td>\n<td align=\"right\">94.1 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">84.9<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.0-8B-Instruct<\/td>\n<td align=\"right\">6.5 %<\/td>\n<td align=\"right\">93.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">74.2<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-1.5-Pro-002<\/td>\n<td align=\"right\">6.6 %<\/td>\n<td align=\"right\">93.7 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">62.0<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-1.5-Flash<\/td>\n<td align=\"right\">6.6 %<\/td>\n<td align=\"right\">93.4 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">63.3<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Pixtral<\/td>\n<td align=\"right\">6.6 %<\/td>\n<td align=\"right\">93.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">76.4<\/td>\n<\/tr>\n<tr>\n<td>Microsoft phi-2<\/td>\n<td align=\"right\">6.7 %<\/td>\n<td align=\"right\">93.3 %<\/td>\n<td align=\"right\">91.5 %<\/td>\n<td align=\"right\">80.8<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-2-2B-it<\/td>\n<td align=\"right\">7.0 %<\/td>\n<td align=\"right\">93.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">62.2<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-3B-Instruct<\/td>\n<td align=\"right\">7.0 %<\/td>\n<td align=\"right\">93.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">70.4<\/td>\n<\/tr>\n<tr>\n<td>Llama-3-8B-Chat-hf<\/td>\n<td align=\"right\">7.4 %<\/td>\n<td align=\"right\">92.6 %<\/td>\n<td align=\"right\">99.8 %<\/td>\n<td align=\"right\">79.7<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Ministral-8B<\/td>\n<td align=\"right\">7.5 %<\/td>\n<td align=\"right\">92.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">62.7<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-Pro<\/td>\n<td align=\"right\">7.7 %<\/td>\n<td align=\"right\">92.3 %<\/td>\n<td align=\"right\">98.4 %<\/td>\n<td align=\"right\">89.5<\/td>\n<\/tr>\n<tr>\n<td>01-AI Yi-1.5-6B-Chat<\/td>\n<td align=\"right\">7.9 %<\/td>\n<td align=\"right\">92.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">98.9<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.2-3B-Instruct<\/td>\n<td align=\"right\">7.9 %<\/td>\n<td align=\"right\">92.1 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">72.2<\/td>\n<\/tr>\n<tr>\n<td>DeepSeek-V3-0324<\/td>\n<td align=\"right\">8.0 %<\/td>\n<td align=\"right\">92.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">78.9<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Ministral-3B<\/td>\n<td align=\"right\">8.3 %<\/td>\n<td align=\"right\">91.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">73.2<\/td>\n<\/tr>\n<tr>\n<td>databricks dbrx-instruct<\/td>\n<td align=\"right\">8.3 %<\/td>\n<td align=\"right\">91.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">85.9<\/td>\n<\/tr>\n<tr>\n<td>Qwen2-VL-2B-Instruct<\/td>\n<td align=\"right\">8.3 %<\/td>\n<td align=\"right\">91.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">81.8<\/td>\n<\/tr>\n<tr>\n<td>Cohere Aya Expanse 32B<\/td>\n<td align=\"right\">8.5 %<\/td>\n<td align=\"right\">91.5 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">81.9<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.1-8B-Instruct<\/td>\n<td align=\"right\">8.6 %<\/td>\n<td align=\"right\">91.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">107.4<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Small2<\/td>\n<td align=\"right\">8.6 %<\/td>\n<td align=\"right\">91.4 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">74.2<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.2-8B-Instruct<\/td>\n<td align=\"right\">8.7 %<\/td>\n<td align=\"right\">91.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">120.1<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.0-2B-Instruct<\/td>\n<td align=\"right\">8.8 %<\/td>\n<td align=\"right\">91.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">81.6<\/td>\n<\/tr>\n<tr>\n<td>Mistral-7B-Instruct-v0.3<\/td>\n<td align=\"right\">9.5 %<\/td>\n<td align=\"right\">90.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">98.4<\/td>\n<\/tr>\n<tr>\n<td>Google Gemini-1.5-Pro<\/td>\n<td align=\"right\">9.1 %<\/td>\n<td align=\"right\">90.9 %<\/td>\n<td align=\"right\">99.8 %<\/td>\n<td align=\"right\">61.6<\/td>\n<\/tr>\n<tr>\n<td>Anthropic Claude-3-opus<\/td>\n<td align=\"right\">10.1 %<\/td>\n<td align=\"right\">89.9 %<\/td>\n<td align=\"right\">95.5 %<\/td>\n<td align=\"right\">92.1<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-2-9B-it<\/td>\n<td align=\"right\">10.1 %<\/td>\n<td align=\"right\">89.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">70.2<\/td>\n<\/tr>\n<tr>\n<td>Llama-2-13B-Chat-hf<\/td>\n<td align=\"right\">10.5 %<\/td>\n<td align=\"right\">89.5 %<\/td>\n<td align=\"right\">99.8 %<\/td>\n<td align=\"right\">82.1<\/td>\n<\/tr>\n<tr>\n<td>AllenAI-OLMo-2-13B-Instruct<\/td>\n<td align=\"right\">10.8 %<\/td>\n<td align=\"right\">89.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">82.0<\/td>\n<\/tr>\n<tr>\n<td>AllenAI-OLMo-2-7B-Instruct<\/td>\n<td align=\"right\">11.1 %<\/td>\n<td align=\"right\">88.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">112.6<\/td>\n<\/tr>\n<tr>\n<td>Mistral-Nemo-Instruct<\/td>\n<td align=\"right\">11.2 %<\/td>\n<td align=\"right\">88.8 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">69.9<\/td>\n<\/tr>\n<tr>\n<td>Llama-2-7B-Chat-hf<\/td>\n<td align=\"right\">11.3 %<\/td>\n<td align=\"right\">88.7 %<\/td>\n<td align=\"right\">99.6 %<\/td>\n<td align=\"right\">119.9<\/td>\n<\/tr>\n<tr>\n<td>Microsoft WizardLM-2-8x22B<\/td>\n<td align=\"right\">11.7 %<\/td>\n<td align=\"right\">88.3 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">140.8<\/td>\n<\/tr>\n<tr>\n<td>Cohere Aya Expanse 8B<\/td>\n<td align=\"right\">12.2 %<\/td>\n<td align=\"right\">87.8 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">83.9<\/td>\n<\/tr>\n<tr>\n<td>Amazon Titan-Express<\/td>\n<td align=\"right\">13.5 %<\/td>\n<td align=\"right\">86.5 %<\/td>\n<td align=\"right\">99.5 %<\/td>\n<td align=\"right\">98.4<\/td>\n<\/tr>\n<tr>\n<td>Google PaLM-2<\/td>\n<td align=\"right\">14.1 %<\/td>\n<td align=\"right\">85.9 %<\/td>\n<td align=\"right\">99.8 %<\/td>\n<td align=\"right\">86.6<\/td>\n<\/tr>\n<tr>\n<td><a href=\"https:\/\/www.kdjingpai.com\/deepseek-r1nenglixiang\/\">DeepSeek-R1<\/a><\/td>\n<td align=\"right\">14.3 %<\/td>\n<td align=\"right\">85.7 %<\/td>\n<td align=\"right\">100.0%<\/td>\n<td align=\"right\">77.1<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-7B-it<\/td>\n<td align=\"right\">14.8 %<\/td>\n<td align=\"right\">85.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">113.0<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.1-2B-Instruct<\/td>\n<td align=\"right\">15.7 %<\/td>\n<td align=\"right\">84.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">107.7<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-1.5B-Instruct<\/td>\n<td align=\"right\">15.8 %<\/td>\n<td align=\"right\">84.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">70.7<\/td>\n<\/tr>\n<tr>\n<td>Qwen-QwQ-32B-Preview<\/td>\n<td align=\"right\">16.1 %<\/td>\n<td align=\"right\">83.9 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">201.5<\/td>\n<\/tr>\n<tr>\n<td>Anthropic Claude-3-sonnet<\/td>\n<td align=\"right\">16.3 %<\/td>\n<td align=\"right\">83.7 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">108.5<\/td>\n<\/tr>\n<tr>\n<td>IBM Granite-3.2-2B-Instruct<\/td>\n<td align=\"right\">16.5 %<\/td>\n<td align=\"right\">83.5 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">117.7<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-1.1-7B-it<\/td>\n<td align=\"right\">17.0 %<\/td>\n<td align=\"right\">83.0 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">64.3<\/td>\n<\/tr>\n<tr>\n<td>Anthropic Claude-2<\/td>\n<td align=\"right\">17.4 %<\/td>\n<td align=\"right\">82.6 %<\/td>\n<td align=\"right\">99.3 %<\/td>\n<td align=\"right\">87.5<\/td>\n<\/tr>\n<tr>\n<td>Google Flan-T5-large<\/td>\n<td align=\"right\">18.3 %<\/td>\n<td align=\"right\">81.7 %<\/td>\n<td align=\"right\">99.3 %<\/td>\n<td align=\"right\">20.9<\/td>\n<\/tr>\n<tr>\n<td>Mixtral-8x7B-Instruct-v0.1<\/td>\n<td align=\"right\">20.1 %<\/td>\n<td align=\"right\">79.9 %<\/td>\n<td align=\"right\">99.9 %<\/td>\n<td align=\"right\">90.7<\/td>\n<\/tr>\n<tr>\n<td>Llama-3.2-1B-Instruct<\/td>\n<td align=\"right\">20.7 %<\/td>\n<td align=\"right\">79.3 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">71.5<\/td>\n<\/tr>\n<tr>\n<td>Apple OpenELM-3B-Instruct<\/td>\n<td align=\"right\">24.8 %<\/td>\n<td align=\"right\">75.2 %<\/td>\n<td align=\"right\">99.3 %<\/td>\n<td align=\"right\">47.2<\/td>\n<\/tr>\n<tr>\n<td>Qwen2.5-0.5B-Instruct<\/td>\n<td align=\"right\">25.2 %<\/td>\n<td align=\"right\">74.8 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">72.6<\/td>\n<\/tr>\n<tr>\n<td>Google Gemma-1.1-2B-it<\/td>\n<td align=\"right\">27.8 %<\/td>\n<td align=\"right\">72.2 %<\/td>\n<td align=\"right\">100.0 %<\/td>\n<td align=\"right\">66.8<\/td>\n<\/tr>\n<tr>\n<td>TII falcon-7B-instruct<\/td>\n<td align=\"right\">29.9 %<\/td>\n<td align=\"right\">70.1 %<\/td>\n<td align=\"right\">90.0 %<\/td>\n<td align=\"right\">75.5<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><em>\u6ce8\uff1a\u6a21\u578b\u6392\u540d\u6839\u636e\u5e7b\u89c9\u7387\u4ece\u4f4e\u5230\u9ad8\u6392\u5e8f\u3002\u5b8c\u6574\u5217\u8868\u548c\u6a21\u578b\u63a5\u5165\u7ec6\u8282\u53ef\u5728\u539f\u59cb HHEM Leaderboard GitHub \u4ed3\u5e93\u67e5\u770b\u3002<\/em><\/p>\n<p>\u89c2\u5bdf\u6392\u884c\u699c\uff0c\u53ef\u4ee5\u770b\u5230 Google \u7684\u00a0<code><a href=\"https:\/\/www.kdjingpai.com\/geminibardgubu\/\">Gemini<\/a><\/code>\u00a0\u7cfb\u5217\u6a21\u578b\u548c OpenAI \u7684\u90e8\u5206\u65b0\u6a21\u578b\uff08\u5982\u00a0<code>o3-mini-high-reasoning<\/code>\uff09\u8868\u73b0\u62a2\u773c\uff0c\u5e7b\u89c9\u7387\u63a7\u5236\u5728\u6781\u4f4e\u7684\u6c34\u5e73\u3002\u8fd9\u663e\u793a\u4e86\u5934\u90e8\u5382\u5546\u5728\u63d0\u5347\u6a21\u578b\u4e8b\u5b9e\u6027\u65b9\u9762\u7684\u8fdb\u5c55\u3002\u540c\u65f6\uff0c\u4e5f\u80fd\u770b\u5230\u4e0d\u540c\u89c4\u6a21\u3001\u4e0d\u540c\u67b6\u6784\u7684\u6a21\u578b\u4e4b\u95f4\u5b58\u5728\u663e\u8457\u5dee\u5f02\u3002\u4e00\u4e9b\u8f83\u5c0f\u7684\u6a21\u578b\uff0c\u5982 Microsoft \u7684\u00a0<code>Phi<\/code>\u00a0\u7cfb\u5217\u6216 Google \u7684\u00a0<code>Gemma<\/code>\u00a0\u7cfb\u5217\uff0c\u4e5f\u53d6\u5f97\u4e86\u4e0d\u9519\u7684\u6210\u7ee9\uff0c\u6697\u793a\u7740\u6a21\u578b\u53c2\u6570\u91cf\u5e76\u975e\u51b3\u5b9a\u4e8b\u5b9e\u4e00\u81f4\u6027\u7684\u552f\u4e00\u56e0\u7d20\u3002\u7136\u800c\uff0c\u4e00\u4e9b\u65e9\u671f\u6216\u7279\u5b9a\u4f18\u5316\u7684\u6a21\u578b\uff0c\u5e7b\u89c9\u7387\u5219\u76f8\u5bf9\u8f83\u9ad8\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u5f3a\u63a8\u7406\u6a21\u578b\u4e0e\u77e5\u8bc6\u5e93\u7684\u201c\u9519\u914d\u201d\uff1a\u4ee5 DeepSeek-R1 \u4e3a\u4f8b<\/h3>\n<p>\u6392\u884c\u699c\u4e2d\u00a0<code>DeepSeek-R1<\/code>\u00a0\u7684\u5e7b\u89c9\u7387\uff0814.3%\uff09\u76f8\u5bf9\u8f83\u9ad8\uff0c\u8fd9\u5f15\u51fa\u4e86\u4e00\u4e2a\u503c\u5f97\u63a2\u8ba8\u7684\u95ee\u9898\uff1a\u4e3a\u4ec0\u4e48\u4e00\u4e9b\u5728\u63a8\u7406\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u8272\u7684\u6a21\u578b\uff0c\u5728\u57fa\u4e8e\u4e8b\u5b9e\u7684\u6458\u8981\u4efb\u52a1\u4e2d\u53cd\u800c\u5bb9\u6613\u4ea7\u751f\u5e7b\u89c9\uff1f<\/p>\n<p><code>DeepSeek-R1<\/code>\u00a0\u8fd9\u7c7b\u6a21\u578b\u901a\u5e38\u88ab\u8bbe\u8ba1\u7528\u4e8e\u5904\u7406\u590d\u6742\u7684\u903b\u8f91\u63a8\u7406\u3001\u9075\u5faa\u6307\u4ee4\u548c\u591a\u6b65\u601d\u8003\u3002\u5b83\u4eec\u7684\u6838\u5fc3\u4f18\u52bf\u5728\u4e8e\u201c\u63a8\u5bfc\u201d\u548c\u201c\u6f14\u7ece\u201d\uff0c\u800c\u975e\u7b80\u5355\u5730\u201c\u590d\u8ff0\u201d\u6216\u201c\u8f6c\u8ff0\u201d\u3002\u7136\u800c\uff0c\u77e5\u8bc6\u5e93\uff08\u5c24\u5176\u662f <a href=\"https:\/\/www.kdjingpai.com\/rag\/\">RAG<\/a> \u573a\u666f\u4e0b\u7684\u77e5\u8bc6\u5e93\uff09\u7684\u6838\u5fc3\u8981\u6c42\u6070\u6070\u662f\u540e\u8005\uff1a\u6a21\u578b\u9700\u8981\u4e25\u683c\u4f9d\u636e\u63d0\u4f9b\u7684\u6587\u672c\u4fe1\u606f\u8fdb\u884c\u56de\u7b54\u6216\u6458\u8981\uff0c\u6700\u5927\u9650\u5ea6\u5730\u907f\u514d\u5f15\u5165\u5916\u90e8\u77e5\u8bc6\u6216\u8fdb\u884c\u8fc7\u5ea6\u5f15\u7533\u3002<\/p>\n<p>\u5f53\u4e00\u4e2a\u5f3a\u63a8\u7406\u6a21\u578b\u88ab\u9650\u5236\u5728\u4ec5\u80fd\u4f7f\u7528\u7ed9\u5b9a\u6587\u6863\u7684\u60c5\u51b5\u4e0b\u8fdb\u884c\u6458\u8981\u65f6\uff0c\u5176\u201c\u63a8\u7406\u201d\u672c\u80fd\u53ef\u80fd\u6210\u4e3a\u4e00\u628a\u53cc\u5203\u5251\u3002\u5b83\u53ef\u80fd\u4f1a\uff1a<\/p>\n<ol>\n<li><strong>\u8fc7\u5ea6\u89e3\u8bfb:<\/strong>\u00a0\u5bf9\u539f\u6587\u4fe1\u606f\u8fdb\u884c\u4e0d\u5fc5\u8981\u7684\u6df1\u5c42\u63a8\u65ad\uff0c\u5f97\u51fa\u539f\u6587\u5e76\u672a\u660e\u793a\u7684\u7ed3\u8bba\u3002<\/li>\n<li><strong>\u7f1d\u5408\u4fe1\u606f:<\/strong>\u00a0\u5c1d\u8bd5\u5c06\u539f\u6587\u788e\u7247\u5316\u7684\u4fe1\u606f\u901a\u8fc7\u201c\u5408\u7406\u201d\u7684\u903b\u8f91\u94fe\u6761\u4e32\u8054\u8d77\u6765\uff0c\u4f46\u8fd9\u4e2a\u94fe\u6761\u53ef\u80fd\u5e76\u975e\u539f\u6587\u6240\u652f\u6301\u3002<\/li>\n<li><strong>\u9ed8\u8ba4\u5916\u90e8\u77e5\u8bc6:<\/strong>\u00a0\u5373\u4fbf\u88ab\u8981\u6c42\u53ea\u4f9d\u636e\u539f\u6587\uff0c\u5176\u8bad\u7ec3\u4e2d\u4e60\u5f97\u7684\u5e9e\u5927\u4e16\u754c\u77e5\u8bc6\u4ecd\u53ef\u80fd\u65e0\u610f\u8bc6\u5730\u6e17\u5165\uff0c\u5bfc\u81f4\u4e0e\u539f\u6587\u4e8b\u5b9e\u7684\u504f\u79bb\u3002<\/li>\n<\/ol>\n<p>\u7b80\u5355\u6765\u8bf4\uff0c\u8fd9\u7c7b\u6a21\u578b\u53ef\u80fd\u201c\u60f3\u5f97\u592a\u591a\u201d\uff0c\u5728\u9700\u8981\u7cbe\u786e\u3001\u5fe0\u5b9e\u590d\u8ff0\u4fe1\u606f\u7684\u573a\u666f\u4e0b\uff0c\u53cd\u800c\u5bb9\u6613\u201c\u806a\u660e\u53cd\u88ab\u806a\u660e\u8bef\u201d\uff0c\u5236\u9020\u51fa\u770b\u4f3c\u5408\u7406\u4f46\u5b9e\u9645\u4e0a\u662f\u5e7b\u89c9\u7684\u5185\u5bb9\u3002\u8fd9\u8bf4\u660e\uff0c\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u548c\u4e8b\u5b9e\u4e00\u81f4\u6027\uff08\u5c24\u5176\u662f\u5728\u53d7\u9650\u4fe1\u606f\u6e90\u4e0b\u7684\u4e8b\u5b9e\u4e00\u81f4\u6027\uff09\u662f\u4e24\u79cd\u4e0d\u540c\u7684\u80fd\u529b\u7ef4\u5ea6\u3002\u9488\u5bf9\u77e5\u8bc6\u5e93\u3001RAG \u7b49\u573a\u666f\uff0c\u9009\u62e9\u5e7b\u89c9\u7387\u4f4e\u3001\u80fd\u5fe0\u5b9e\u53cd\u6620\u8f93\u5165\u4fe1\u606f\u7684\u6a21\u578b\uff0c\u53ef\u80fd\u6bd4\u5355\u7eaf\u8ffd\u6c42\u63a8\u7406\u5f97\u5206\u66f4\u91cd\u8981\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u65b9\u6cd5\u8bba\u4e0e\u80cc\u666f<\/h3>\n<p>HHEM \u6392\u884c\u699c\u7684\u5efa\u7acb\u5e76\u975e\u51ed\u7a7a\u800c\u6765\uff0c\u5b83\u501f\u9274\u4e86\u4e8b\u5b9e\u4e00\u81f4\u6027\u7814\u7a76\u9886\u57df\u7684\u8bf8\u591a\u5148\u524d\u5de5\u4f5c\uff0c\u5982\u00a0<code>SUMMAC<\/code>,\u00a0<code>TRUE<\/code>,\u00a0<code>TrueTeacher<\/code>\u00a0\u7b49\u8bba\u6587\u4e2d\u5efa\u7acb\u7684\u65b9\u6cd5\u8bba\u3002\u5176\u6838\u5fc3\u601d\u8def\u662f\u8bad\u7ec3\u4e00\u4e2a\u4e13\u95e8\u7528\u4e8e\u68c0\u6d4b\u5e7b\u89c9\u7684\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u5728\u5224\u65ad\u6458\u8981\u4e0e\u539f\u6587\u4e00\u81f4\u6027\u65b9\u9762\u8fbe\u5230\u4e0e\u4eba\u7c7b\u8bc4\u4f30\u8005\u9ad8\u5ea6\u76f8\u5173\u7684\u6c34\u5e73\u3002<\/p>\n<p>\u8bc4\u4f30\u8fc7\u7a0b\u9009\u53d6\u4e86\u6458\u8981\u4efb\u52a1\u4f5c\u4e3a LLM \u4e8b\u5b9e\u6027\u7684\u4ee3\u8868\u3002\u8fd9\u4e0d\u4ec5\u56e0\u4e3a\u6458\u8981\u4efb\u52a1\u672c\u8eab\u8981\u6c42\u9ad8\u5ea6\u7684\u4e8b\u5b9e\u4e00\u81f4\u6027\uff0c\u4e5f\u56e0\u4e3a\u5b83\u4e0e RAG \u7cfb\u7edf\u7684\u5de5\u4f5c\u6a21\u5f0f\u9ad8\u5ea6\u76f8\u4f3c\u2014\u2014\u5728 RAG \u4e2d\uff0cLLM \u6b63\u662f\u626e\u6f14\u4e86\u5bf9\u68c0\u7d22\u5230\u7684\u4fe1\u606f\u8fdb\u884c\u6574\u5408\u4e0e\u6458\u8981\u7684\u89d2\u8272\u3002\u56e0\u6b64\uff0c\u8fd9\u4e2a\u6392\u884c\u699c\u7684\u7ed3\u679c\u5bf9\u4e8e\u8bc4\u4f30\u6a21\u578b\u5728 RAG \u5e94\u7528\u4e2d\u7684\u53ef\u9760\u6027\u5177\u6709\u53c2\u8003\u4ef7\u503c\u3002<\/p>\n<p>\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u8bc4\u4f30\u56e2\u961f\u6392\u9664\u4e86\u90a3\u4e9b\u6a21\u578b\u62d2\u7edd\u56de\u7b54\u6216\u7ed9\u51fa\u6781\u77ed\u65e0\u6548\u7b54\u6848\u7684\u6587\u6863\uff0c\u6700\u7ec8\u4f7f\u7528\u4e86\u6240\u6709\u6a21\u578b\u90fd\u80fd\u6210\u529f\u751f\u6210\u6458\u8981\u7684 831 \u4efd\u6587\u6863\uff08\u6e90\u81ea\u6700\u521d\u7684 1006 \u4efd\uff09\u8fdb\u884c\u6700\u7ec8\u6392\u540d\u8ba1\u7b97\uff0c\u4ee5\u786e\u4fdd\u516c\u5e73\u6027\u3002\u56de\u7b54\u7387\u548c\u5e73\u5747\u6458\u8981\u957f\u5ea6\u6307\u6807\u4e5f\u53cd\u6620\u4e86\u6a21\u578b\u5728\u5904\u7406\u8fd9\u4e9b\u8bf7\u6c42\u65f6\u7684\u884c\u4e3a\u6a21\u5f0f\u3002<\/p>\n<p>\u8bc4\u4f30\u4f7f\u7528\u7684 Prompt \u6a21\u677f\u5982\u4e0b\uff1a<\/p>\n<pre><code>You are a chat bot answering questions using data. You must stick to the answers provided solely by the text in the passage provided. You are asked the question 'Provide a concise summary of the following passage, covering the core <a href=\"https:\/\/www.kdjingpai.com\/pieces-for-developers\/\">pieces<\/a> of information described.' &lt;PASSAGE&gt;'\r\n<\/code><\/pre>\n<p>\u5728\u5b9e\u9645\u8c03\u7528\u65f6\uff0c<code>&lt;PASSAGE&gt;<\/code>\u00a0\u4f1a\u88ab\u66ff\u6362\u4e3a\u5177\u4f53\u7684\u6e90\u6587\u6863\u5185\u5bb9\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u5c55\u671b\u672a\u6765<\/h3>\n<p>HHEM \u6392\u884c\u699c\u9879\u76ee\u65b9\u8868\u793a\uff0c\u672a\u6765\u8ba1\u5212\u6269\u5c55\u8bc4\u4f30\u8303\u56f4\uff1a<\/p>\n<ul>\n<li><strong>\u5f15\u7528\u51c6\u786e\u6027:<\/strong>\u00a0\u589e\u52a0\u5bf9 LLM \u5728 RAG \u573a\u666f\u4e0b\u5f15\u7528\u6765\u6e90\u51c6\u786e\u6027\u7684\u8bc4\u4f30\u3002<\/li>\n<li><strong>\u5176\u4ed6 RAG \u4efb\u52a1:<\/strong>\u00a0\u8986\u76d6\u66f4\u591a RAG \u76f8\u5173\u4efb\u52a1\uff0c\u4f8b\u5982\u591a\u6587\u6863\u6458\u8981\u3002<\/li>\n<li><strong>\u591a\u8bed\u8a00\u652f\u6301:<\/strong>\u00a0\u5c06\u8bc4\u4f30\u6269\u5c55\u5230\u82f1\u8bed\u4e4b\u5916\u7684\u5176\u4ed6\u8bed\u8a00\u3002<\/li>\n<\/ul>\n<p>HHEM \u6392\u884c\u699c\u4e3a\u89c2\u5bdf\u548c\u6bd4\u8f83\u4e0d\u540c LLM \u5728\u63a7\u5236\u5e7b\u89c9\u3001\u4fdd\u6301\u4e8b\u5b9e\u4e00\u81f4\u6027\u65b9\u9762\u7684\u80fd\u529b\u63d0\u4f9b\u4e86\u4e00\u4e2a\u6709\u4ef7\u503c\u7684\u7a97\u53e3\u3002\u867d\u7136\u5b83\u5e76\u975e\u8861\u91cf\u6a21\u578b\u8d28\u91cf\u7684\u552f\u4e00\u6807\u51c6\uff0c\u4e5f\u65e0\u6cd5\u6db5\u76d6\u6240\u6709\u7c7b\u578b\u7684\u5e7b\u89c9\uff0c\u4f46\u5b83\u65e0\u7591\u63a8\u52a8\u4e86\u884c\u4e1a\u5bf9 LLM \u53ef\u9760\u6027\u95ee\u9898\u7684\u5173\u6ce8\uff0c\u5e76\u4e3a\u5f00\u53d1\u8005\u9009\u62e9\u548c\u4f18\u5316\u6a21\u578b\u63d0\u4f9b\u4e86\u91cd\u8981\u7684\u53c2\u8003\u4f9d\u636e\u3002\u968f\u7740\u6a21\u578b\u548c\u8bc4\u4f30\u65b9\u6cd5\u7684\u6301\u7eed\u8fed\u4ee3\uff0c\u6211\u4eec\u6709\u671b\u770b\u5230 LLM \u5728\u63d0\u4f9b\u51c6\u786e\u3001\u53ef\u4fe1\u4fe1\u606f\u65b9\u9762\u53d6\u5f97\u66f4\u5927\u8fdb\u6b65\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u80fd\u529b\u65e5\u65b0\u6708\u5f02\uff0c\u4f46\u5176\u8f93\u51fa\u5185\u5bb9\u4e2d\u51fa\u73b0\u4e8b\u5b9e\u6027\u9519\u8bef\u6216\u4e0e\u539f\u6587\u65e0\u5173\u4fe1\u606f\u7684\u201c\u5e7b\u89c9\u201d\u73b0\u8c61\uff0c\u59cb\u7ec8\u662f\u963b\u788d\u5176\u66f4\u5e7f\u6cdb\u5e94\u7528\u548c\u6df1\u5ea6\u4fe1\u4efb\u7684\u4e00\u5927\u96be\u9898\u3002\u4e3a\u4e86\u91cf\u5316\u8bc4\u4f30\u8fd9\u4e00\u95ee\u9898\uff0cHughes Hallucination Evaluation Model &#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[46],"tags":[],"class_list":["post-29641","post","type-post","status-publish","format-standard","hentry","category-news"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/29641","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=29641"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/29641\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=29641"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=29641"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=29641"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}