{"id":37596,"date":"2025-01-31T18:01:02","date_gmt":"2025-01-31T10:01:02","guid":{"rendered":"https:\/\/17aitech.com\/?p=37596"},"modified":"2025-01-31T18:01:02","modified_gmt":"2025-01-31T10:01:02","slug":"%e6%97%b6%e9%9a%946%e5%b9%b4%ef%bc%8c%e8%b0%b7%e6%ad%8cbert%e7%bb%88%e4%ba%8e%e6%9c%89%e6%9b%bf%e4%bb%a3%e5%93%81%e4%ba%86%ef%bc%81%e6%9b%b4%e5%bf%ab%e6%9b%b4%e5%87%86%e6%9b%b4%e9%95%bf%ef%bc%8c","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=37596","title":{"rendered":"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-01-05-2\" target=\"_blank\">\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI<\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"11\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>\u771f\u6b63\u6709\u7528\u7684\u4e3b\u529b\u6a21\u578b\u3002<\/p>\n<\/section>\n<\/blockquote>\n<p>BERT \u4e8e 2018 \u5e74\u53d1\u5e03\uff0c\u8fd9\u4e2a\u65f6\u95f4\u70b9\uff0c\u5982\u679c\u6309\u7167 AI \u7684\u7eaa\u4e8b\u5e74\u4ee3\u6765\u8bf4\uff0c\u53ef\u4ee5\u8bf4\u662f\u4e00\u5343\u5e74\u4ee5\u524d\u7684\u4e8b\u4e86\uff01\u867d\u7136\u8fc7\u53bb\u4e86\u8fd9\u4e48\u591a\u5e74\uff0c\u4f46\u81f3\u4eca\u4ecd\u88ab\u5e7f\u6cdb\u4f7f\u7528\uff1a\u4e8b\u5b9e\u4e0a\uff0c\u5b83\u76ee\u524d\u662f HuggingFace \u4e2d\u5fc3\u4e0b\u8f7d\u91cf\u7b2c\u4e8c\u5927\u7684\u6a21\u578b\uff0c\u6bcf\u6708\u4e0b\u8f7d\u91cf\u8d85\u8fc7 6800 \u4e07\u6b21\u3002<\/p>\n<p>\u597d\u6d88\u606f\u662f\uff0c\u516d\u5e74\u540e\u7684\u4eca\u5929\uff0c\u6211\u4eec\u7ec8\u4e8e\u6709\u4e86\u66ff\u4ee3\u54c1\uff01<\/p>\n<p>\u8fd1\u65e5\uff0c\u65b0\u578b AI \u7814\u53d1\u5b9e\u9a8c\u5ba4 Answer.AI\u3001\u82f1\u4f1f\u8fbe\u7b49\u53d1\u5e03\u4e86 ModernBERT\u3002<\/p>\n<p>ModernBERT \u662f\u4e00\u4e2a\u65b0\u7684\u6a21\u578b\u7cfb\u5217\uff0c\u5177\u6709\u4e24\u4e2a\u578b\u53f7\uff1a\u57fa\u7840\u7248 139M \u3001\u8f83\u5927\u7248 395M\u3002\u5728\u901f\u5ea6\u548c\u51c6\u786e\u7387\u65b9\u9762\u90fd\u6bd4 BERT \u53ca\u5176\u540c\u7c7b\u6a21\u578b\u6709\u4e86\u663e\u8457\u6539\u8fdb\u3002\u8be5\u6a21\u578b\u91c7\u7528\u4e86\u8fd1\u5e74\u6765\u5728\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u65b9\u9762\u7684\u6570\u5341\u9879\u8fdb\u5c55\uff0c\u5305\u62ec\u5bf9\u67b6\u6784\u548c\u8bad\u7ec3\u8fc7\u7a0b\u7684\u66f4\u65b0\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png\"><\/a><\/p>\n<p>\u9664\u4e86\u901f\u5ea6\u66f4\u5feb\u3001\u51c6\u786e\u5ea6\u66f4\u9ad8\u5916\uff0cModernBERT \u8fd8\u5c06\u4e0a\u4e0b\u6587\u957f\u5ea6\u589e\u52a0\u5230 8k \u4e2a token\uff0c\u800c\u5927\u591a\u6570\u7f16\u7801\u5668\u53ea\u6709 512 \u4e2a token\uff0c\u5e76\u4e14\u662f\u7b2c\u4e00\u4e2a\u5728\u8bad\u7ec3\u6570\u636e\u4e2d\u5305\u542b\u5927\u91cf\u4ee3\u7801\u7684\u4ec5\u7f16\u7801\u5668\u4e13\u7528\u6a21\u578b\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-79f510196b387f1cb32fe8c5ffc88ec5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-79f510196b387f1cb32fe8c5ffc88ec5.png\"><\/a><\/p>\n<ul>\n<li>\n<section>\u539f\u6587\u5730\u5740\uff1ahttps:\/\/huggingface.co\/blog\/modernbert<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2412.13663<\/section>\n<\/li>\n<li>\n<section>\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/github.com\/huggingface\/blog\/blob\/main\/modernbert.md<\/section>\n<\/li>\n<\/ul>\n<p>Answer.AI \u8054\u5408\u521b\u59cb\u4eba Jeremy Howard \u8868\u793a\uff0cModernBERT \u4e0d\u7092\u4f5c\u751f\u6210\u5f0f AI\uff08GenAI\uff09\uff0c\u800c\u662f\u771f\u6b63\u7684\u4e3b\u529b\u6a21\u578b\uff0c\u53ef\u7528\u4e8e\u68c0\u7d22\u3001\u5206\u7c7b\uff0c\u662f\u771f\u6b63\u6709\u7528\u7684\u5de5\u4f5c\u3002\u6b64\u5916\u8fd8\u66f4\u5feb\u3001\u66f4\u51c6\u786e\u3001\u4e0a\u4e0b\u6587\u66f4\u957f\u3001\u66f4\u6709\u7528\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-83495ecea0ac8cb07de1aca6037900bf.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-83495ecea0ac8cb07de1aca6037900bf.png\"><\/a><\/p>\n<p><strong>\u4ec5\u89e3\u7801\u5668\u6a21\u578b<\/strong><\/p>\n<p>\u8fd1\u671f\uff0cGPT\u3001Llama \u548c Claude \u7b49\u6a21\u578b\u8fc5\u901f\u5d1b\u8d77\uff0c\u8fd9\u4e9b\u6a21\u578b\u662f\u4ec5\u89e3\u7801\u5668\u6a21\u578b\uff0c\u4e5f\u53ef\u4ee5\u8bf4\u662f\u751f\u6210\u6a21\u578b\u3002\u5b83\u4eec\u7684\u51fa\u73b0\u50ac\u751f\u51fa\u4ee4\u4eba\u60ca\u53f9\u7684\u65b0\u578b\u9886\u57df GenAI\uff0c\u4f8b\u5982\u751f\u6210\u827a\u672f\u548c\u4ea4\u4e92\u5f0f\u804a\u5929\u3002\u672c\u6587\u6240\u505a\u7684\uff0c\u672c\u8d28\u4e0a\u5c31\u662f\u5c06\u8fd9\u4e9b\u8fdb\u5c55\u79fb\u690d\u56de\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u3002<\/p>\n<p>\u4e3a\u4ec0\u4e48\u8981\u8fd9\u4e48\u505a\uff1f\u56e0\u4e3a\u8bb8\u591a\u5b9e\u9645\u5e94\u7528\u90fd\u9700\u8981\u4e00\u4e2a\u7cbe\u7b80\u7684\u6a21\u578b\uff01\u800c\u4e14\u5b83\u4e0d\u9700\u8981\u662f\u4e00\u4e2a\u751f\u6210\u6a21\u578b\u3002<\/p>\n<p>\u66f4\u76f4\u767d\u5730\u8bf4\uff0c\u4ec5\u89e3\u7801\u5668\u6a21\u578b\u5bf9\u4e8e\u8bb8\u591a\u5de5\u4f5c\u6765\u8bf4\u592a\u5927\u3001\u592a\u6162\u3001\u592a\u79c1\u5bc6\u3001\u592a\u6602\u8d35\u3002\u60f3\u60f3\u6700\u521d\u7684 GPT-1 \u662f\u4e00\u4e2a 1.17 \u4ebf\u53c2\u6570\u7684\u6a21\u578b\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cLlama 3.1 \u6a21\u578b\u6709 4050 \u4ebf\u4e2a\u53c2\u6570\uff0c\u5bf9\u4e8e\u5927\u591a\u6570\u516c\u53f8\u6765\u8bf4\uff0c\u8fd9\u79cd\u65b9\u6cd5\u8fc7\u4e8e\u590d\u6742\u548c\u6602\u8d35\uff0c\u65e0\u6cd5\u590d\u5236\u3002<\/p>\n<p>GenAI \u7684\u6d41\u884c\u70ed\u6f6e\u63a9\u76d6\u4e86\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u7684\u4f5c\u7528\u3002\u8fd9\u4e9b\u6a21\u578b\u5728\u8bb8\u591a\u79d1\u5b66\u548c\u5546\u4e1a\u5e94\u7528\u4e2d\u53d1\u6325\u4e86\u5de8\u5927\u4f5c\u7528\u3002<\/p>\n<p><strong>\u4ec5\u7f16\u7801\u5668\u6a21\u578b<\/strong><\/p>\n<p>\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u7684\u8f93\u51fa\u662f\u6570\u503c\u5217\u8868\uff08\u5d4c\u5165\u5411\u91cf\uff09\uff0c\u5176\u5c06\u300c\u7b54\u6848\u300d\u76f4\u63a5\u7f16\u7801\u4e3a\u538b\u7f29\u7684\u6570\u5b57\u5f62\u5f0f\u3002\u8be5\u5411\u91cf\u662f\u6a21\u578b\u8f93\u5165\u7684\u538b\u7f29\u8868\u793a\uff0c\u8fd9\u5c31\u662f\u4e3a\u4ec0\u4e48\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u6709\u65f6\u88ab\u79f0\u4e3a\u8868\u793a\u6a21\u578b\u3002<\/p>\n<p>\u867d\u7136\u4ec5\u89e3\u7801\u5668\u6a21\u578b\uff08\u5982 GPT\uff09\u53ef\u4ee5\u5b8c\u6210\u4ec5\u7f16\u7801\u5668\u6a21\u578b\uff08\u5982 BERT\uff09\u7684\u5de5\u4f5c\uff0c\u4f46\u5b83\u4eec\u53d7\u5230\u4e00\u4e2a\u5173\u952e\u7ea6\u675f\uff1a\u7531\u4e8e\u5b83\u4eec\u662f\u751f\u6210\u6a21\u578b\uff0c\u56e0\u6b64\u4ece\u6570\u5b66\u4e0a\u8bb2\u5b83\u4eec\u300c\u4e0d\u5141\u8bb8\u7aa5\u89c6\u300d\u540e\u9762\u7684 token\u3002\u8fd9\u4e0e\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u5f62\u6210\u5bf9\u6bd4\uff0c\u540e\u8005\u7ecf\u8fc7\u8bad\u7ec3\uff0c\u6bcf\u4e2a token \u90fd\u53ef\u4ee5\u5411\u524d\u548c\u5411\u540e\uff08\u53cc\u5411\uff09\u67e5\u770b\u3002\u5b83\u4eec\u5c31\u662f\u4e3a\u6b64\u800c\u6784\u5efa\u7684\uff0c\u8fd9\u4f7f\u5f97\u5b83\u4eec\u5728\u5de5\u4f5c\u4e2d\u975e\u5e38\u9ad8\u6548\u3002<\/p>\n<p>\u57fa\u672c\u4e0a\uff0c\u50cf OpenAI \u00a0o1 \u8fd9\u6837\u7684\u524d\u6cbf\u6a21\u578b\u5c31\u50cf\u6cd5\u62c9\u5229 SF-23\uff0c\u8fd9\u663e\u7136\u662f\u5de5\u7a0b\u5b66\u7684\u80dc\u5229\uff0c\u65e8\u5728\u8d62\u5f97\u6bd4\u8d5b\u3002\u4f46\u4ec5\u662f\u66f4\u6362\u8f6e\u80ce\u5c31\u9700\u8981\u4e13\u95e8\u7684\u7ef4\u4fee\u7ad9\uff0c\u800c\u4e14\u4f60\u65e0\u6cd5\u81ea\u5df1\u8d2d\u4e70\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cBERT \u6a21\u578b\u5c31\u50cf\u672c\u7530\u601d\u57df\uff0c\u8fd9\u4e5f\u662f\u4e00\u9879\u5de5\u7a0b\u5b66\u7684\u80dc\u5229\uff0c\u4f46\u66f4\u5fae\u5999\u7684\u662f\uff0c\u5b83\u88ab\u8bbe\u8ba1\u6210\u4ef7\u683c\u5b9e\u60e0\u3001\u7701\u6cb9\u3001\u53ef\u9760\u4e14\u6781\u5176\u5b9e\u7528\u7684\u8f66\u578b\u3002\u56e0\u800c\u65e0\u5904\u4e0d\u5728\u3002<\/p>\n<p>\u6211\u4eec\u53ef\u4ee5\u4ece\u4e0d\u540c\u7684\u89d2\u5ea6\u770b\u5f85\u8fd9\u4e2a\u95ee\u9898\u3002<\/p>\n<p>\u9996\u5148\u662f\u57fa\u4e8e\u7f16\u7801\u5668\u7684\u7cfb\u7edf\uff1a\u5728 GPT \u51fa\u73b0\u4e4b\u524d\uff0c\u793e\u4ea4\u5a92\u4f53\u548c Netflix \u7b49\u5e73\u53f0\u4e2d\u90fd\u6709\u5185\u5bb9\u63a8\u8350\u3002\u8fd9\u4e9b\u7cfb\u7edf\u4e0d\u662f\u5efa\u7acb\u5728\u751f\u6210\u6a21\u578b\u4e0a\uff0c\u800c\u662f\u5efa\u7acb\u5728\u8868\u5f81\u6a21\u578b\uff08\u5982\u4ec5\u7f16\u7801\u5668\u6a21\u578b\uff09\u4e0a\u3002\u6240\u6709\u8fd9\u4e9b\u7cfb\u7edf\u4ecd\u7136\u5b58\u5728\uff0c\u5e76\u4e14\u4ecd\u5728\u5927\u89c4\u6a21\u8fd0\u884c\u3002<\/p>\n<p>\u63a5\u7740\u662f\u4e0b\u8f7d\u91cf\uff1a\u5728 HuggingFace \u4e0a\uff0cRoBERTa \u662f\u57fa\u4e8e BERT \u7684\u9886\u5148\u6a21\u578b\u4e4b\u4e00\uff0c\u5176\u4e0b\u8f7d\u91cf\u8d85\u8fc7\u4e86 HuggingFace \u4e0a 10 \u4e2a\u6700\u53d7\u6b22\u8fce\u7684 LLM \u7684\u603b\u4e0b\u8f7d\u91cf\u3002\u4e8b\u5b9e\u4e0a\uff0c\u76ee\u524d\uff0c\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u6bcf\u6708\u7684\u4e0b\u8f7d\u91cf\u603b\u8ba1\u8d85\u8fc7 10 \u4ebf\u6b21\uff0c\u51e0\u4e4e\u662f\u4ec5\u89e3\u7801\u5668\u6a21\u578b\uff08\u6bcf\u6708\u4e0b\u8f7d\u91cf\u4e3a 3.97 \u4ebf\u6b21\uff09\u7684\u4e09\u500d\u3002<\/p>\n<p>\u6700\u540e\u662f\u63a8\u7406\u6210\u672c\uff1a\u6309\u63a8\u7406\u8ba1\u7b97\uff0c\u6bcf\u5e74\u5728\u7eaf\u7f16\u7801\u5668\u6a21\u578b\u4e0a\u6267\u884c\u7684\u63a8\u7406\u8981\u6bd4\u7eaf\u89e3\u7801\u5668\u6a21\u578b\u6216\u751f\u6210\u6a21\u578b\u591a\u5f88\u591a\u500d\u3002\u4e00\u4e2a\u6709\u8da3\u7684\u4f8b\u5b50\u662f FineWeb-Edu\uff0c\u56e2\u961f\u9009\u62e9\u4f7f\u7528\u4ec5\u89e3\u7801\u5668\u6a21\u578b Llama-3-70b-Instruct \u751f\u6210\u6ce8\u91ca\uff0c\u5e76\u4f7f\u7528\u7ecf\u8fc7\u5fae\u8c03\u7684\u57fa\u4e8e BERT \u7684\u6a21\u578b\u6267\u884c\u5927\u90e8\u5206\u8fc7\u6ee4\u3002\u6b64\u8fc7\u6ee4\u8017\u65f6 6,000 H100 \u5c0f\u65f6\uff0c\u6309\u7167\u6bcf\u5c0f\u65f6 10 \u7f8e\u5143\u7684\u5b9a\u4ef7\uff0c\u603b\u8ba1 60,000 \u7f8e\u5143\u3002\u53e6\u4e00\u65b9\u9762\uff0c\u5373\u4f7f\u4f7f\u7528\u6210\u672c\u6700\u4f4e\u7684 Google Gemini Flash \u53ca\u5176\u4f4e\u63a8\u7406\u6210\u672c\uff08\u6bcf\u767e\u4e07\u4e2a token 0.075 \u7f8e\u5143\uff09\uff0c\u5411\u6d41\u884c\u7684\u4ec5\u89e3\u7801\u5668\u6a21\u578b\u63d0\u4f9b 15 \u4e07\u4ebf\u4e2a token \u4e5f\u9700\u8981\u82b1\u8d39\u8d85\u8fc7\u4e00\u767e\u4e07\u7f8e\u5143\uff01<\/p>\n<p><strong>\u6027\u80fd<\/strong><\/p>\n<p>\u4ee5\u4e0b\u662f ModernBERT \u548c\u5176\u4ed6\u6a21\u578b\u5728\u4e00\u7cfb\u5217\u4efb\u52a1\u4e2d\u7684\u51c6\u786e\u7387\u6bd4\u8f83 \u3002ModernBERT \u662f\u552f\u4e00\u5728\u6bcf\u4e2a\u7c7b\u522b\u4e2d\u90fd\u83b7\u5f97\u6700\u9ad8\u5206\u7684\u6a21\u578b\uff1a<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-4885372822a00a5f4354a07638781307.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-4885372822a00a5f4354a07638781307.png\"><\/a><\/p>\n<p>\u5982\u679c\u4f60\u4e86\u89e3 Kaggle \u4e0a\u7684 NLP \u7ade\u8d5b\uff0c\u90a3\u4e48\u4f60\u5c31\u4f1a\u77e5\u9053 DeBERTaV3 \u591a\u5e74\u6765\u4e00\u76f4\u662f\u5904\u4e8e\u51a0\u519b\u4f4d\u7f6e\u3002\u4f46\u73b0\u5728\u60c5\u51b5\u5df2\u7ecf\u53d8\u4e86\uff1aModernBERT \u4e0d\u4ec5\u662f\u7b2c\u4e00\u4e2a\u5728 GLUE \u4e0a\u51fb\u8d25 DeBERTaV3 \u7684\u6a21\u578b\uff0c\u800c\u4e14\u5b83\u4f7f\u7528\u7684\u5185\u5b58\u8fd8\u4e0d\u5230 Deberta \u7684 1\/5\u3002<\/p>\n<p>\u5f53\u7136\uff0cModernBERT \u7684\u901f\u5ea6\u4e5f\u5f88\u5feb\uff0c\u5b83\u7684\u901f\u5ea6\u662f DeBERTa \u7684\u4e24\u500d \u2014\u2014 \u4e8b\u5b9e\u4e0a\uff0c\u5728\u8f93\u5165\u6df7\u5408\u957f\u5ea6\u7684\u60c5\u51b5\u4e0b\uff0c\u901f\u5ea6\u6700\u9ad8\u53ef\u63d0\u9ad8 4 \u500d\u3002\u5b83\u7684\u957f\u4e0a\u4e0b\u6587\u63a8\u7406\u901f\u5ea6\u6bd4\u5176\u4ed6\u9ad8\u8d28\u91cf\u6a21\u578b\uff08\u5982 NomicBERT \u548c GTE-en-MLM\uff09\u5feb\u8fd1 3 \u500d\u3002<\/p>\n<p>ModernBERT \u7684\u4e0a\u4e0b\u6587\u957f\u5ea6\u4e3a 8,192 \u4e2a token\uff0c\u6bd4\u5927\u591a\u6570\u73b0\u6709\u7f16\u7801\u5668\u957f 16 \u500d\u4ee5\u4e0a\u3002<\/p>\n<p>\u5bf9\u4e8e\u4ee3\u7801\u68c0\u7d22\uff0cModernBERT \u7684\u6027\u80fd\u662f\u72ec\u4e00\u65e0\u4e8c\u7684\uff0c\u56e0\u4e3a\u4e4b\u524d\u4ece\u672a\u6709\u8fc7\u7f16\u7801\u5668\u6a21\u578b\u5728\u5927\u91cf\u4ee3\u7801\u6570\u636e\u4e0a\u7ecf\u8fc7\u8bad\u7ec3\u3002\u4f8b\u5982\uff0c\u5728 StackOverflow-QA \u6570\u636e\u96c6 (SQA) \u4e0a\uff0c\u8fd9\u662f\u4e00\u4e2a\u6df7\u5408\u4e86\u4ee3\u7801\u548c\u81ea\u7136\u8bed\u8a00\u7684\u6df7\u5408\u6570\u636e\u96c6\uff0cModernBERT \u7684\u4e13\u4e1a\u4ee3\u7801\u7406\u89e3\u548c\u957f\u4e0a\u4e0b\u6587\u4f7f\u5176\u6210\u4e3a\u552f\u4e00\u4e00\u4e2a\u5728\u6b64\u4efb\u52a1\u4e0a\u5f97\u5206\u8d85\u8fc7 80 \u7684\u6a21\u578b\u3002<\/p>\n<p>\u4e0e\u4e3b\u6d41\u6a21\u578b\u76f8\u6bd4\uff0cModernBERT \u5728\u68c0\u7d22\u3001\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u548c\u4ee3\u7801\u68c0\u7d22\u8fd9\u4e09\u5927\u7c7b\u4efb\u52a1\u4e2d\u8868\u73b0\u7a81\u51fa\u3002\u867d\u7136 ModernBERT \u5728\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\u4e0a\u7a0d\u5fae\u843d\u540e\u4e8e DeBERTaV3\uff0c\u4f46\u5176\u901f\u5ea6\u8981\u5feb\u5f88\u591a\u500d\u3002<\/p>\n<p>\u4e0e\u7279\u5b9a\u9886\u57df\u7684\u6a21\u578b\u76f8\u6bd4\uff0cModernBERT \u5728\u5927\u591a\u6570\u4efb\u52a1\u4e2d\u90fd\u76f8\u5f53\u6216\u66f4\u80dc\u4e00\u7b79\u3002\u6b64\u5916\uff0cModernBERT \u5728\u5927\u591a\u6570\u4efb\u52a1\u4e2d\u90fd\u6bd4\u5927\u591a\u6570\u6a21\u578b\u66f4\u5feb\uff0c\u5e76\u4e14\u53ef\u4ee5\u5904\u7406\u591a\u8fbe 8,192 \u4e2a token \u7684\u8f93\u5165\uff0c\u6bd4\u4e3b\u6d41\u6a21\u578b\u957f 16 \u500d\u3002<\/p>\n<p><strong>\u6548\u7387<\/strong><\/p>\n<p>\u4ee5\u4e0b\u662f ModernBERT \u548c\u5176\u4ed6\u89e3\u7801\u5668\u6a21\u578b\u5728 NVIDIA RTX 4090 \u4e0a\u7684\u5185\u5b58\uff08\u6700\u5927\u6279\u5927\u5c0f\uff0cBS\uff09\u548c\u63a8\u7406\uff08\u4ee5\u6bcf\u79d2\u6570\u5343\u4e2a token \u4e3a\u5355\u4f4d\uff09\u6548\u7387\u5bf9\u6bd4\u7ed3\u679c\uff1a<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-070e1971b246b340e207b7f821d699ec.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-070e1971b246b340e207b7f821d699ec.png\"><\/a><\/p>\n<p>\u5bf9\u4e8e\u53ef\u53d8\u957f\u5ea6\u7684\u8f93\u5165\uff0cModernBERT \u6bd4\u5176\u4ed6\u6a21\u578b\u90fd\u5feb\u5f97\u591a\u3002<\/p>\n<p>\u5bf9\u4e8e\u957f\u4e0a\u4e0b\u6587\u8f93\u5165\uff0cModernBERT \u6bd4\u7b2c\u4e8c\u5feb\u7684\u6a21\u578b\u5feb 2-3 \u500d\u3002\u6b64\u5916\uff0c\u7531\u4e8e ModernBERT \u7684\u6548\u7387\uff0c\u5b83\u53ef\u4ee5\u4f7f\u7528\u6bd4\u51e0\u4e4e\u4efb\u4f55\u5176\u4ed6\u6a21\u578b\u90fd\u5927\u7684\u6279\u5904\u7406\u5927\u5c0f\uff0c\u5e76\u4e14\u53ef\u4ee5\u5728\u66f4\u5c0f\u3001\u66f4\u4fbf\u5b9c\u7684 GPU \u4e0a\u6709\u6548\u4f7f\u7528\u3002\u7279\u522b\u662f\u57fa\u7840\u6a21\u578b\u7684\u6548\u7387\u53ef\u80fd\u4f1a\u4f7f\u65b0\u5e94\u7528\u7a0b\u5e8f\u80fd\u591f\u76f4\u63a5\u5728\u6d4f\u89c8\u5668\u3001\u624b\u673a\u7b49\u4e0a\u8fd0\u884c\u3002<\/p>\n<p><strong>ModernBERT<\/strong><\/p>\n<p>\u63a5\u4e0b\u6765\u6587\u7ae0\u89e3\u91ca\u4e86\u4e3a\u4ec0\u4e48\u6211\u4eec\u5e94\u8be5\u66f4\u52a0\u91cd\u89c6\u7f16\u7801\u5668\u6a21\u578b\u3002\u4f5c\u4e3a\u503c\u5f97\u4fe1\u8d56\u3001\u4f46\u88ab\u4f4e\u4f30\u7684\u4e3b\u529b\u6a21\u578b\uff0c\u81ea 2018 \u5e74 BERT \u53d1\u5e03\u4ee5\u6765\uff0c\u5b83\u4eec\u7684\u66f4\u65b0\u901f\u5ea6\u51fa\u5947\u5730\u6162\uff01<\/p>\n<p>\u66f4\u4ee4\u4eba\u60ca\u8bb6\u7684\u662f\uff1a\u81ea RoBERTa \u4ee5\u6765\uff0c\u8fd8\u6ca1\u6709\u7f16\u7801\u5668\u80fd\u591f\u63d0\u4f9b\u6574\u4f53\u7684\u6539\u8fdb\uff1aDeBERTaV3 \u5177\u6709\u66f4\u597d\u7684 GLUE \u548c\u5206\u7c7b\u6027\u80fd\uff0c\u4f46\u727a\u7272\u4e86\u6548\u7387\u548c\u68c0\u7d22\u3002\u5176\u4ed6\u6a21\u578b\uff08\u4f8b\u5982 AlBERT\uff09\u6216\u8f83\u65b0\u7684\u6a21\u578b\uff08\u4f8b\u5982 GTE-en-MLM\uff09\u90fd\u5728\u67d0\u4e9b\u65b9\u9762\u6bd4\u539f\u59cb BERT \u548c RoBERTa \u6709\u6240\u6539\u8fdb\uff0c\u4f46\u5728\u5176\u4ed6\u65b9\u9762\u6709\u6240\u9000\u6b65\u3002<\/p>\n<p>ModernBERT \u9879\u76ee\u4e3b\u8981\u6709\u4e09\u4e2a\u6838\u5fc3\u70b9\uff1a<\/p>\n<ul>\n<li>\n<section>\u73b0\u4ee3\u5316\u7684 transformer \u67b6\u6784\uff1b<\/section>\n<\/li>\n<li>\n<section>\u7279\u522b\u91cd\u89c6\u6ce8\u610f\u529b\u6548\u7387\uff1b<\/section>\n<\/li>\n<li>\n<section>\u4ee5\u53ca\u6570\u636e\u3002<\/section>\n<\/li>\n<\/ul>\n<p><strong>\u8ba4\u8bc6\u65b0\u7684 Transformer<\/strong><\/p>\n<p>Transformer \u67b6\u6784\u5df2\u6210\u4e3a\u4e3b\u6d41\uff0c\u5982\u4eca\u7edd\u5927\u591a\u6570\u6a21\u578b\u90fd\u5728\u4f7f\u7528\u5b83\u3002\u4f46\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cTransformer \u4e0d\u6b62\u4e00\u4e2a\uff0c\u800c\u662f\u6709\u5f88\u591a\u79cd\u3002\u5176\u4e3b\u8981\u5171\u540c\u70b9\u662f\uff0c\u90fd\u575a\u4fe1\u6ce8\u610f\u529b\u673a\u5236\u624d\u662f\u6240\u9700\u8981\u7684\u4e00\u5207\uff0c\u56e0\u6b64\u56f4\u7ed5\u6ce8\u610f\u529b\u673a\u5236\u6784\u5efa\u4e86\u5404\u79cd\u6539\u8fdb\u3002<\/p>\n<p>\u7528 RoPE \u66ff\u6362\u65e7\u7684\u4f4d\u7f6e\u7f16\u7801\uff1a\u8fd9\u4f7f\u5f97\u6a21\u578b\u80fd\u591f\u66f4\u597d\u5730\u7406\u89e3\u5355\u8bcd\u4e4b\u95f4\u7684\u5173\u7cfb\uff0c\u5e76\u5141\u8bb8\u6269\u5c55\u5230\u66f4\u957f\u7684\u5e8f\u5217\u957f\u5ea6\u3002<\/p>\n<ul>\n<li>\n<section>\u5c06\u65e7\u7684 MLP \u5c42\u66ff\u6362\u4e3a GeGLU \u5c42\uff0c\u6539\u8fdb\u539f\u59cb BERT \u7684 GeLU \u6fc0\u6d3b\u51fd\u6570\uff1b<\/section>\n<\/li>\n<li>\n<section>\u901a\u8fc7\u5220\u9664\u4e0d\u5fc5\u8981\u7684\u504f\u7f6e\u9879\u6765\u7b80\u5316\u67b6\u6784\uff0c\u4ece\u800c\u66f4\u6709\u6548\u5730\u4f7f\u7528\u53c2\u6570\u9884\u7b97\uff1b<\/section>\n<\/li>\n<li>\n<section>\u5728\u5d4c\u5165\u540e\u589e\u52a0\u4e00\u4e2a\u989d\u5916\u7684\u5f52\u4e00\u5316\u5c42\uff0c\u6709\u52a9\u4e8e\u7a33\u5b9a\u8bad\u7ec3\u3002<\/section>\n<\/li>\n<\/ul>\n<p><strong>\u5168\u5c40\u548c\u5c40\u90e8\u6ce8\u610f\u529b<\/strong><\/p>\n<p>ModernBERT \u6700\u5177\u5f71\u54cd\u529b\u7684\u529f\u80fd\u4e4b\u4e00\u662f Alternating \u6ce8\u610f\u529b\u673a\u5236\uff0c\u800c\u4e0d\u662f\u5168\u5c40\u6ce8\u610f\u529b\u673a\u5236\u3002<\/p>\n<p>\u968f\u7740\u6ce8\u610f\u529b\u8ba1\u7b97\u590d\u6742\u5ea6\u968f\u7740\u6bcf\u4e2a\u989d\u5916\u7684 token \u800c\u81a8\u80c0\uff0c\u8fd9\u610f\u5473\u7740 ModernBERT \u53ef\u4ee5\u6bd4\u4efb\u4f55\u5176\u4ed6\u6a21\u578b\u66f4\u5feb\u5730\u5904\u7406\u957f\u8f93\u5165\u5e8f\u5217\u3002\u5b9e\u9645\u4e0a\uff0c\u5b83\u770b\u8d77\u6765\u50cf\u8fd9\u6837\uff1a<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-8c0f33aa3a8ba932b6e351f3e9011070.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-8c0f33aa3a8ba932b6e351f3e9011070.png\"><\/a><\/p>\n<p><strong>Unpadding \u548c\u5e8f\u5217 Packing<\/strong><\/p>\n<p>\u53e6\u4e00\u4e2a\u6709\u52a9\u4e8e ModernBERT \u63d0\u9ad8\u6548\u7387\u7684\u6838\u5fc3\u673a\u5236\u662f Unpadding \u548c\u5e8f\u5217 Packing\u3002<\/p>\n<p>\u4e3a\u4e86\u80fd\u591f\u5904\u7406\u540c\u4e00\u6279\u6b21\u4e2d\u7684\u591a\u4e2a\u5e8f\u5217\uff0c\u7f16\u7801\u5668\u6a21\u578b\u8981\u6c42\u5e8f\u5217\u5177\u6709\u76f8\u540c\u7684\u957f\u5ea6\uff0c\u4ee5\u4fbf\u53ef\u4ee5\u6267\u884c\u5e76\u884c\u8ba1\u7b97\u3002\u4f20\u7edf\u4e0a\u7684\u505a\u6cd5\u662f\u4f9d\u9760 padding \u6765\u5b9e\u73b0\u8fd9\u4e00\u70b9\uff1a\u627e\u51fa\u54ea\u4e2a\u53e5\u5b50\u6700\u957f\uff0c\u5e76\u6dfb\u52a0\u65e0\u610f\u4e49\u7684 token \u6765\u586b\u5145\u5e8f\u5217\u3002<\/p>\n<p>\u867d\u7136\u586b\u5145\u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898\uff0c\u4f46\u5b83\u5e76\u4e0d\u4f18\u96c5\uff1a\u5927\u91cf\u7684\u8ba1\u7b97\u6700\u7ec8\u88ab\u6d6a\u8d39\u5728\u586b\u5145 token \u4e0a\uff0c\u800c\u8fd9\u4e9b token \u4e0d\u4f1a\u63d0\u4f9b\u4efb\u4f55\u8bed\u4e49\u4fe1\u606f\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-e5cb52e330c9ae30c776aabefa9fddf1.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-e5cb52e330c9ae30c776aabefa9fddf1.png\"><\/a><\/p>\n<p><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u6bd4\u8f83 padding \u548c\u5e8f\u5217 packing\u3002\u5e8f\u5217 packing\uff08unpadding \uff09\u907f\u514d\u4e86\u6a21\u578b\u5728\u586b\u5145 token \u4e0a\u7684\u8ba1\u7b97\u6d6a\u8d39\u3002<\/sup><\/em><\/p>\n<p>Unpadding \u89e3\u51b3\u4e86\u8fd9\u4e2a\u95ee\u9898\uff0c\u5176\u4e0d\u4f1a\u4fdd\u7559\u8fd9\u4e9b\u586b\u5145 token\uff0c\u800c\u662f\u5c06\u5b83\u4eec\u5168\u90e8\u5220\u9664\uff0c\u5e76\u5c06\u5b83\u4eec\u8fde\u63a5\u5230\u6279\u5927\u5c0f\u4e3a 1 \u7684\u5c0f\u6279\u6b21\u4e2d\uff0c\u4ece\u800c\u907f\u514d\u4e86\u4e0d\u5fc5\u8981\u7684\u8ba1\u7b97\u3002\u5982\u679c\u4f60\u4f7f\u7528\u7684\u662f Flash Attention\uff0cunpadding \u751a\u81f3\u6bd4\u4ee5\u524d\u7684\u65b9\u6cd5\u66f4\u5feb\uff0c\u6bd4\u4ee5\u524d\u7684\u65b9\u6cd5\u52a0\u5feb 10-20%\u3002<\/p>\n<p><strong>\u8bad\u7ec3<\/strong><\/p>\n<p>\u7f16\u7801\u5668\u843d\u540e\u7684\u4e00\u5927\u65b9\u9762\u5728\u4e8e\u8bad\u7ec3\u6570\u636e\u3002\u5728\u901a\u5e38\u7684\u7406\u89e3\u4e2d\uff0c\u8fd9\u4ec5\u6307\u7684\u662f\u8bad\u7ec3\u6570\u636e\u89c4\u6a21\uff0c\u4f46\u4e8b\u5b9e\u5e76\u975e\u5982\u6b64\uff1a\u4ee5\u5f80\u7684\u7f16\u7801\u5668\uff08\u5982 DeBERTaV3\uff09\u7ecf\u8fc7\u8db3\u591f\u957f\u65f6\u95f4\u7684\u8bad\u7ec3\uff0c\u751a\u81f3\u53ef\u80fd\u7a81\u7834\u4e86\u4e07\u4ebf tokens \u89c4\u6a21\u3002<\/p>\n<p>\u4f46\u95ee\u9898\u5728\u4e8e\u6570\u636e\u591a\u6837\u6027\uff1a\u5f88\u591a\u65e7\u6a21\u578b\u5728\u6709\u9650\u7684\u8bed\u6599\u5e93\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u901a\u5e38\u5305\u62ec Wikipedia \u548c Wikibooks\u3002\u5f88\u660e\u663e\uff0c\u8fd9\u4e9b\u6df7\u5408\u6570\u636e\u662f\u5355\u4e00\u7684\u6587\u672c\u6a21\u6001\uff1a\u5b83\u4eec\u53ea\u5305\u542b\u9ad8\u8d28\u91cf\u7684\u81ea\u7136\u6587\u672c\u3002<\/p>\n<p>\u76f8\u6bd4\u4e4b\u4e0b\uff0cModernBERT \u7684\u8bad\u7ec3\u6570\u636e\u5177\u6709\u591a\u79cd\u82f1\u8bed\u6765\u6e90\uff0c\u5305\u62ec\u7f51\u9875\u6587\u6863\u3001\u4ee3\u7801\u548c\u79d1\u5b66\u6587\u7ae0\u3002\u8be5\u6a21\u578b\u8bad\u7ec3\u4e86 2 \u4e07\u4ebf tokens\uff0c\u5176\u4e2d\u5927\u591a\u6570\u662f\u552f\u4e00\u7684\uff0c\u800c\u4e0d\u50cf\u4ee5\u5f80\u7f16\u7801\u5668\u90a3\u6837\u5e38\u5e38\u91cd\u590d 20 \u5230 40 \u6b21\u3002\u8fd9\u79cd\u505a\u6cd5\u7684\u5f71\u54cd\u662f\u663e\u800c\u6613\u89c1\u7684\uff1a\u5728\u6240\u6709\u7684\u5f00\u6e90\u7f16\u7801\u5668\u4e2d\uff0cModernBERT \u5728\u7f16\u7801\u4efb\u52a1\u4e0a\u5b9e\u73b0\u4e86 SOTA\u3002<\/p>\n<p><strong>\u6d41\u7a0b<\/strong><\/p>\n<p>\u56e2\u961f\u575a\u6301\u539f\u59cb BERT \u8bad\u7ec3\u65b9\u6cd5\uff0c\u5e76\u5728\u540e\u7eed\u5de5\u4f5c\u7684\u542f\u53d1\u4e0b\u8fdb\u884c\u4e86\u4e00\u4e9b\u5c0f\u7684\u5347\u7ea7\uff0c\u5305\u62ec\u5220\u9664\u4e86\u4e0b\u4e00\u53e5\uff08Next-Sentence\uff09\u9884\u6d4b\u76ee\u6807\uff0c\u539f\u56e0\u662f\u5b83\u5728\u589e\u52a0\u5f00\u9500\u7684\u60c5\u51b5\u4e0b\u6ca1\u6709\u660e\u663e\u7684\u6536\u76ca\uff0c\u5e76\u5c06\u63a9\u853d\u7387\u4ece 15% \u63d0\u9ad8\u4e86 30%\u3002<\/p>\n<p>\u4e24\u4e2a\u6a21\u578b\u90fd\u91c7\u7528\u4e09\u6bb5\u5f0f\u8bad\u7ec3\u6d41\u7a0b\u3002\u9996\u5148\u5728\u5e8f\u5217\u957f\u5ea6\u4e3a 1024 \u7684\u60c5\u51b5\u4e0b\u8bad\u7ec3\u4e86 1.7T tokens \u7684\u6570\u636e\uff0c\u7136\u540e\u91c7\u7528\u4e00\u4e2a\u957f\u4e0a\u4e0b\u6587\u9002\u5e94\u9636\u6bb5\uff0c\u5728\u5e8f\u5217\u957f\u5ea6\u4e3a 8192 \u7684\u60c5\u51b5\u4e0b\u8bad\u7ec3\u4e86 250B tokens \u7684\u6570\u636e\uff0c\u540c\u65f6\u901a\u8fc7\u964d\u4f4e\u6279\u5927\u5c0f\u6765\u4fdd\u6301\u6bcf\u4e2a\u6279\u6b21\u7684\u603b tokens \u6570\u5927\u4f53\u4e00\u81f4\u3002\u6700\u540e\u6309\u7167 ProLong \u4e2d\u5f3a\u8c03\u7684\u957f\u4e0a\u4e0b\u6587\u6269\u5c55\u601d\u8def\uff0c\u5bf9\u4e0d\u540c\u91c7\u6837\u7684 50B tokens \u6570\u636e\u8fdb\u884c\u9000\u706b\u3002<\/p>\n<p>\u4ece\u7ed3\u679c\u6765\u770b\uff0c\u4e09\u6bb5\u5f0f\u8bad\u7ec3\u53ef\u4ee5\u786e\u4fdd\u6a21\u578b\u8fd0\u884c\u826f\u597d\uff0c\u5728\u957f\u4e0a\u4e0b\u6587\u4efb\u52a1\u4e0a\u5177\u6709\u7ade\u4e89\u529b\uff0c\u4e5f\u4e0d\u4f1a\u5f71\u54cd\u5176\u5904\u7406\u77ed\u4e0a\u4e0b\u6587\u7684\u80fd\u529b\u3002<\/p>\n<p>\u6b64\u5916\u8fd8\u6709\u53e6\u4e00\u4e2a\u597d\u5904\uff1a\u5bf9\u4e8e\u524d\u4e24\u4e2a\u9636\u6bb5\uff0c\u56e2\u961f\u5728\u9884\u70ed\u9636\u6bb5\u5b8c\u6210\u4e4b\u540e\u4f7f\u7528\u6052\u5b9a\u5b66\u4e60\u7387\u6765\u8bad\u7ec3\uff0c\u53ea\u5bf9\u6700\u540e\u7684 50B tokens \u6267\u884c\u5b66\u4e60\u7387\u8870\u51cf\uff0c\u5e76\u9075\u5faa\u4e86\u68af\u5f62\uff08\u9884\u70ed &#8211; \u7a33\u5b9a &#8211; \u8870\u51cf\uff09\u5b66\u4e60\u7387\u3002\u66f4\u91cd\u8981\u7684\u662f\uff1a\u53d7\u5230 Pythia \u7684\u542f\u53d1\uff0c\u56e2\u961f\u53ef\u4ee5\u5728\u8fd9\u4e9b\u7a33\u5b9a\u7684\u9636\u6bb5\u53d1\u5e03\u6bcf\u4e00\u4e2a\u76f4\u63a5\u4e2d\u95f4\u68c0\u67e5\u70b9\u3002\u8fd9\u6837\u505a\u7684\u4e3b\u8981\u539f\u56e0\u662f\u652f\u6301\u672a\u6765\u7684\u7814\u7a76\u548c\u5e94\u7528\uff1a\u4efb\u4f55\u4eba\u90fd\u53ef\u4ee5\u4ece\u56e2\u961f\u7684\u9884\u8870\u51cf\u68c0\u67e5\u70b9\u91cd\u65b0\u5f00\u59cb\u8bad\u7ec3\uff0c\u5e76\u5bf9\u9002\u5408\u81ea\u5df1\u9884\u671f\u7528\u9014\u7684\u57df\u6570\u636e\u8fdb\u884c\u9000\u706b\u3002<\/p>\n<p><strong>\u6280\u5de7<\/strong><\/p>\n<p>\u6700\u540e\uff0c\u8be5\u56e2\u961f\u4f7f\u7528\u4e86\u4e24\u4e2a\u6280\u5de7\u6765\u52a0\u5feb\u5b9e\u73b0\u901f\u5ea6\u3002<\/p>\n<p>\u7b2c\u4e00\u4e2a\u6280\u5de7\u5f88\u5e38\u89c1\uff1a\u7531\u4e8e\u521d\u59cb\u8bad\u7ec3\u6b65\u9aa4\u66f4\u65b0\u4e86\u968f\u673a\u6743\u91cd\uff0c\u56e0\u800c\u91c7\u7528\u6279\u5927\u5c0f\u9884\u70ed\u3002\u9996\u5148\u4ece\u8f83\u5c0f\u7684\u6279\u5927\u5c0f\u5f00\u59cb\uff0c\u8fd9\u6837\u76f8\u540c\u6570\u91cf\u7684 token \u4f1a\u66f4\u52a0\u9891\u7e41\u5730\u66f4\u65b0\u6a21\u578b\u6743\u91cd\uff0c\u63a5\u4e0b\u6765\u9010\u6e10\u589e\u52a0\u6279\u5927\u5c0f\u76f4\u5230\u8fbe\u5230\u6700\u7ec8\u7684\u8bad\u7ec3\u5927\u5c0f\u3002\u8fd9\u6837\u505a\u5927\u5927\u52a0\u5feb\u4e86\u521d\u59cb\u9636\u6bb5\u7684\u6a21\u578b\u8bad\u7ec3\u901f\u5ea6\uff0c\u5176\u4e2d\u6a21\u578b\u5b66\u4e60\u5230\u4e86\u6700\u57fa\u7840\u7684\u8bed\u8a00\u7406\u89e3\u3002<\/p>\n<p>\u7b2c\u4e8c\u4e2a\u6280\u5de7\u4e0d\u592a\u5e38\u89c1\uff1a\u901a\u8fc7\u5bf9\u8f83\u5927\u7684\u6a21\u578b\u5e73\u94fa\uff08tiling\uff09\u6765\u8fdb\u884c\u6743\u91cd\u521d\u59cb\u5316\uff0c\u8fd9\u4e00\u7075\u611f\u6765\u81ea\u5fae\u8f6f\u7684 Phi \u7cfb\u5217\u6a21\u578b\u3002\u8be5\u6280\u5de7\u57fa\u4e8e\u4ee5\u4e0b\u8ba4\u77e5\uff1a\u5f53\u6211\u4eec\u6709\u4e00\u7ec4\u975e\u5e38\u597d\u7684 ModernBERT-base \u6743\u91cd\u65f6\uff0c\u4e3a\u4ec0\u4e48\u8981\u5229\u7528\u968f\u673a\u6570\u6765\u521d\u59cb\u5316 ModernBERT-large \u7684\u521d\u59cb\u6743\u91cd\u5462\uff1f\u7684\u786e\uff0c\u4e8b\u5b9e\u8bc1\u660e\u4e86\uff0c\u5c06 ModernBERT-base \u7684\u6743\u91cd\u5e73\u94fa\u5230 ModernBERT-large \u8981\u6bd4\u968f\u673a\u6743\u91cd\u7684\u521d\u59cb\u5316\u6548\u679c\u597d\u3002\u6b64\u5916\uff0c\u8be5\u6280\u5de7\u8ddf\u6279\u5927\u5c0f\u9884\u70ed\u4e00\u6837\u53ef\u4ee5\u5728\u5806\u53e0\u540e\u5e26\u6765\u989d\u5916\u597d\u5904\uff0c\u4ece\u800c\u66f4\u5feb\u5730\u8fdb\u884c\u521d\u59cb\u8bad\u7ec3\u3002<\/p>\n<p>\u603b\u4e4b\uff0cModernBERT \u6210\u4e3a\u65b0\u7684\u5c0f\u578b\u3001\u9ad8\u6548\u7684\u4ec5\u7f16\u7801\u5668 SOTA \u7cfb\u5217\u6a21\u578b\uff0c\u5e76\u4e3a BERT \u63d0\u4f9b\u4e86\u4e9f\u9700\u7684\u91cd\u505a\u3002\u53e6\u5916\u8bc1\u660e\u4e86\u4ec5\u7f16\u7801\u5668\u6a21\u578b\u53ef\u4ee5\u901a\u8fc7\u73b0\u4ee3\u65b9\u6cd5\u5f97\u5230\u6539\u8fdb\uff0c\u5e76\u5728\u4e00\u4e9b\u4efb\u52a1\u4e0a\u4ecd\u80fd\u63d0\u4f9b\u975e\u5e38\u5f3a\u5927\u7684\u6027\u80fd\uff0c\u5e76\u5b9e\u73b0\u6781\u5177\u5438\u5f15\u529b\u7684\u5c3a\u5bf8 \/ \u6027\u80fd\u6bd4\u3002<\/p>\n<p><em><sup>\u53c2\u8003\u5185\u5bb9\uff1a<\/sup><\/em><\/p>\n<p><em><sup>https:\/\/huggingface.co\/blog\/modernbert<\/sup><\/em><\/p>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-01-05-2\" target=\"_blank\">\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[81],"class_list":["post-37596","post","type-post","status-publish","format-standard","hentry","category-news","tag-81"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI - \u4e00\u8d77AI\u6280\u672f<\/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:\/\/17aitech.com\/?p=37596\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=37596\",\"url\":\"https:\/\/17aitech.com\/?p=37596\",\"name\":\"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=37596#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=37596#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png\",\"datePublished\":\"2025-01-31T10:01:02+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=37596#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=37596\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=37596#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=37596#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI - \u4e00\u8d77AI\u6280\u672f","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:\/\/17aitech.com\/?p=37596","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=37596","url":"https:\/\/17aitech.com\/?p=37596","name":"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=37596#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=37596#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png","datePublished":"2025-01-31T10:01:02+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=37596#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=37596"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=37596#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/01\/frc-9449900ef627540349a6a4bb5007d1c2.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=37596#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u65f6\u96946\u5e74\uff0c\u8c37\u6b4cBERT\u7ec8\u4e8e\u6709\u66ff\u4ee3\u54c1\u4e86\uff01\u66f4\u5feb\u66f4\u51c6\u66f4\u957f\uff0c\u8fd8\u4e0d\u7092\u4f5cGenAI"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/37596","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=37596"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/37596\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=37596"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=37596"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=37596"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}