{"id":30309,"date":"2026-10-09T11:07:27","date_gmt":"2026-10-09T03:07:27","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=30309"},"modified":"2026-10-09T11:14:46","modified_gmt":"2026-10-09T03:14:46","slug":"google%e6%96%b0%e7%a0%94%e7%a9%b6%ef%bd%9c%e5%8a%a8%e6%80%81%e5%a4%9a%e6%99%ba%e8%83%bd%e4%bd%93%ef%bc%8c%e6%af%8f%e4%b8%aa%e6%9f%a5%e8%af%a2%e8%87%aa%e5%8a%a8%e6%90%adagent%e7%b3%bb%e7%bb%9f","status":"publish","type":"post","link":"https:\/\/www.ixpower.cn\/index.php\/2026\/10\/09\/30309\/","title":{"rendered":"Google\u65b0\u7814\u7a76\uff5c\u52a8\u6001\u591a\u667a\u80fd\u4f53\uff0c\u6bcf\u4e2a\u67e5\u8be2\u81ea\u52a8\u642dAgent\u7cfb\u7edf"},"content":{"rendered":"<div class=\"article-desc\">SHIFT \u5c06\u591a\u667a\u80fd\u4f53\u7cfb\u7edf\u7684\u6784\u5efa\u4ece\u4f9d\u8d56\u7ecf\u9a8c\u4e0e\u8bd5\u9519\uff0c\u63a8\u8fdb\u5230\u53ef\u5b66\u4e60\u3001\u53ef\u9884\u6d4b\u7684\u6846\u67b6\u3002\u5b83\u4ee5\u51bb\u7ed3\u9aa8\u5e72\u3001\u4ec5\u8bad\u7ec3 LoRA \u7684\u8f7b\u91cf Gemma \u4f5c\u4e3a\u67b6\u6784\u5e08\uff0c\u628a\u8bbe\u8ba1\u6210\u672c\u4ece\u6bcf\u6b21\u63a8\u7406\u8f6c\u79fb\u5230\u4e00\u6b21\u6027\u8bad\u7ec3\u5e76\u644a\u8584\uff0c\u4e0d\u4f9d\u8d56\u66f4\u5927\u7684\u6267\u884c\u5668\uff0c\u5728\u516d\u4e2a\u57fa\u51c6\u4e0a\u540c\u65f6\u53d6\u5f97\u6700\u9ad8\u5e73\u5747\u51c6\u786e\u7387\u4e0e\u663e\u8457\u66f4\u4f4e\u7684\u6210\u672c\u6863\u4f4d\uff0c\u53ef\u76f4\u63a5\u7528\u4e8e\u667a\u80fd\u4f53\u4ea7\u54c1\u3002<\/div>\n<div id=\"postspictures\" class=\"article-content\">\n<div id=\"container\" class=\"container am-engine\" data-v-01a18e2f=\"\" data-element=\"root\">\n<p>\u4e00\u79cd\u5e38\u89c1\u7684\u591a\u667a\u80fd\u4f53\u914d\u7f6e\u662f planner\u3001solver\u3001verifier \u5404\u53f8\u5176\u804c\uff0c\u5de5\u5177\u5168\u5f00\uff0c\u5bf9\u6240\u6709\u8bf7\u6c42\u5957\u7528\u540c\u4e00\u5957\u7ed3\u6784\u3002\u67e5\u4e00\u4e2a\u62a5\u8868\u6570\u503c\u4e0e\u6838\u5bf9\u4e00\u6574\u672c\u9884\u7b97\u8868\uff0c\u8d70\u7684\u662f\u540c\u4e00\u6761\u6d41\u7a0b\u3002Google \u4e0e\u4e9a\u5229\u6851\u90a3\u5dde\u7acb\u5927\u5b66\u63d0\u51fa\u7684 SHIFT \u9488\u5bf9\u8fd9\u4e00\u505a\u6cd5\u7ed9\u51fa\u66ff\u4ee3\u65b9\u6848\uff0c\u5373\u6309\u67e5\u8be2\u52a8\u6001\u6784\u5efa harness\uff0c\u7b80\u5355\u4efb\u52a1\u4ec5\u4fdd\u7559\u5355\u4e2a reader\uff0c\u590d\u6742\u4efb\u52a1\u518d\u5f15\u5165\u9a8c\u8bc1\u56de\u8def\u4e0e Python \u6267\u884c\u3002<\/p>\n<p>&nbsp;<\/p>\n<p><img data-dominant-color=\"ded9d9\" data-has-transparency=\"false\" style=\"--dominant-color: #ded9d9;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-30313\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/9543b8990a68f1d327c408c5ad7781ca8298b4.jpg\" width=\"900\" height=\"176\" alt=\"fig0_cover.png\" title=\"fig0_cover.png\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/9543b8990a68f1d327c408c5ad7781ca8298b4.jpg 900w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/9543b8990a68f1d327c408c5ad7781ca8298b4-300x59.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/9543b8990a68f1d327c408c5ad7781ca8298b4-150x29.jpg 150w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/9543b8990a68f1d327c408c5ad7781ca8298b4-768x150.jpg 768w\" sizes=\"auto, (max-width: 900px) 100vw, 900px\" \/><span class=\"data-image-title\">fig0_cover.png<\/span><\/p>\n<p>&nbsp;<\/p>\n<h4>\u4e00\u3001\u95ee\u9898\u4e0e\u52a8\u673a<\/h4>\n<p>\u8bba\u6587\u5c06\u667a\u80fd\u4f53\u7cfb\u7edf\u7684\u89d2\u8272\u3001\u6307\u4ee4\u3001\u5de5\u5177\u4e0e\u901a\u4fe1\u7ed3\u6784\u7edf\u79f0\u4e3a harness\u3002\u5176\u96be\u70b9\u5728\u4e8e\uff0c\u4e00\u79cd\u8bbe\u8ba1\u7684\u4f18\u52a3\u53ea\u80fd\u901a\u8fc7\u5b9e\u9645\u6267\u884c\u89c2\u6d4b\uff0c\u800c\u6267\u884c\u672c\u8eab\u4ea7\u751f\u6210\u672c\u3002\u82e5\u4e3a\u6bcf\u4e2a\u67e5\u8be2\u8bc4\u4f30\u591a\u79cd\u5019\u9009\u7ed3\u6784\uff0c\u5f00\u9500\u5c06\u6210\u500d\u653e\u5927\u3002<\/p>\n<p>&nbsp;<\/p>\n<p><img data-dominant-color=\"dde0df\" data-has-transparency=\"false\" style=\"--dominant-color: #dde0df;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-30314\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/c5a66175742d537b76d9460155686e45ceb1e4.jpg\" width=\"900\" height=\"225\" alt=\"fig1_contrast.png\" title=\"fig1_contrast.png\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/c5a66175742d537b76d9460155686e45ceb1e4.jpg 900w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/c5a66175742d537b76d9460155686e45ceb1e4-300x75.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/c5a66175742d537b76d9460155686e45ceb1e4-150x38.jpg 150w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/c5a66175742d537b76d9460155686e45ceb1e4-768x192.jpg 768w\" sizes=\"auto, (max-width: 900px) 100vw, 900px\" \/><span class=\"data-image-title\">fig1_contrast.png<\/span><\/p>\n<p>&nbsp;<\/p>\n<p>Figure 1 \u5c55\u793a\u4e86\u8fd9\u4e00\u5dee\u5f02\u3002\u5de6\u4fa7\u7684\u56fa\u5b9a harness \u4ee5\u540c\u4e00 Reader \u4e0e Solver \u5904\u7406\u4e09\u9053\u4efb\u52a1\uff0c\u7b80\u5355\u67e5\u8be2\u5c1a\u53ef\uff0c\u4f46\u5217\u5408\u8ba1\u8ba1\u7b97\u5f97\u5230 190\u3001\u589e\u957f\u7387\u8ba1\u7b97\u5f97\u5230 10%\uff0c\u4e24\u5904\u5747\u9519\u8bef\u3002\u53f3\u4fa7 SHIFT \u6309\u67e5\u8be2\u6784\u5efa\u7ed3\u6784\uff0c\u7b80\u5355\u67e5\u8be2\u4ec5\u7528\u4e00\u4e2a reader\uff0c\u5408\u8ba1\u4efb\u52a1\u914d\u7f6e\u9a8c\u8bc1\u56de\u8def\uff0c\u589e\u957f\u7387\u4efb\u52a1\u914d\u7f6e\u5e26 Python \u7684\u9a8c\u8bc1\u5668\uff0c\u4e09\u9053\u5747\u6b63\u786e\u3002<\/p>\n<h4>\u4e8c\u3001\u65b9\u6cd5<\/h4>\n<p>SHIFT \u7684\u6838\u5fc3\u662f\u4e00\u4e2a\u8f7b\u91cf\u672c\u5730\u6a21\u578b\uff0c\u4f5c\u8005\u79f0\u4e3a architect\uff0c\u57fa\u4e8e Gemma \u9aa8\u5e72\uff0c\u63a5\u4e24\u4e2a\u8f93\u51fa\u5934\u3002policy \u5934\u9884\u6d4b\u4e0b\u4e00\u6b65\u7684\u6784\u5efa\u52a8\u4f5c\uff0cvalue \u5934\u5bf9\u5c1a\u672a\u6267\u884c\u7684\u7ed3\u6784\u76f4\u63a5\u4f30\u8ba1\u6548\u7528\uff0c\u8be5\u6548\u7528\u540c\u65f6\u6743\u8861\u51c6\u786e\u7387\u4e0e\u6267\u884c\u6210\u672c\u3002<\/p>\n<p>&nbsp;<\/p>\n<p><img data-dominant-color=\"e0e1de\" data-has-transparency=\"false\" style=\"--dominant-color: #e0e1de;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-30315\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/455ba7f46c5394a5ed28268def16fbfcd1b472.jpg\" width=\"900\" height=\"213\" alt=\"fig2_pipeline.png\" title=\"fig2_pipeline.png\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/455ba7f46c5394a5ed28268def16fbfcd1b472.jpg 900w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/455ba7f46c5394a5ed28268def16fbfcd1b472-300x71.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/455ba7f46c5394a5ed28268def16fbfcd1b472-150x36.jpg 150w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/455ba7f46c5394a5ed28268def16fbfcd1b472-768x182.jpg 768w\" sizes=\"auto, (max-width: 900px) 100vw, 900px\" \/><span class=\"data-image-title\">fig2_pipeline.png<\/span><\/p>\n<p>&nbsp;<\/p>\n<p>\u7ea2\u6846\u6807\u6ce8\u4e24\u5904\u5173\u952e\u7ed3\u6784\uff0c\u5de6\u4fa7\u4e3a policy \u4e0e value \u4e24\u4e2a\u5934\uff0c\u53f3\u4fa7\u4e3a Execute\uff0c\u4ec5\u88ab\u9009\u4e2d\u7684\u7ed3\u6784\u4f1a\u771f\u6b63\u8c03\u7528\u6267\u884c\u5668\u3002\u641c\u7d22\u9636\u6bb5\u91c7\u7528\u8499\u7279\u5361\u6d1b\u6811\u641c\u7d22\uff0c\u4f9d\u636e\u67b6\u6784\u5e08\u7684\u9884\u6d4b\u7ec4\u5408\u89d2\u8272\u3001\u6307\u4ee4\u4e0e\u5de5\u5177\uff0c\u5168\u7a0b\u4e0d\u8c03\u7528\u6267\u884c\u5668\u3002\u6784\u5efa\u8fc7\u7a0b\u5728\u5355\u5757 H100 \u4e0a\u8017\u65f6\u4e0d\u8db3\u4e00\u79d2\uff0c\u76f8\u5bf9\u901a\u5e38\u6301\u7eed\u6570\u5341\u79d2\u81f3\u6570\u5206\u949f\u7684\u6267\u884c\uff0c\u989d\u5916\u5f00\u9500\u53ef\u5ffd\u7565\u3002\u8bad\u7ec3\u9636\u6bb5\u624d\u5b9e\u9645\u6267\u884c\uff0cpolicy \u4ece\u641c\u7d22\u8bbf\u95ee\u8ba1\u6570\u4e2d\u5b66\u4e60\uff0cvalue \u4ece\u6267\u884c\u7ed3\u679c\u4e2d\u5b66\u4e60\uff0c\u5956\u52b1\u4e3a\u4efb\u52a1\u6210\u529f\u51cf\u53bb token\u3001\u5ef6\u8fdf\u3001\u8d85\u65f6\u4ee5\u53ca\u65b0\u589e\u5de5\u5177\u4e0e\u667a\u80fd\u4f53\u7684\u60e9\u7f5a\u3002\u8be5\u8bad\u7ec3\u6210\u672c\u4e00\u6b21\u6027\u6295\u5165\uff0c\u5e76\u5728\u540e\u7eed\u67e5\u8be2\u4e0a\u644a\u8584\u3002<\/p>\n<h4>\u4e09\u3001\u5173\u952e\u524d\u63d0\uff0c\u4e0d\u5b58\u5728\u901a\u7528 harness<\/h4>\n<p>\u8be5\u5de5\u4f5c\u7684\u4e00\u4e2a\u91cd\u8981\u8bba\u636e\u662f\uff0c\u7edf\u4e00\u914d\u7f6e\u5728\u591a\u6570\u573a\u666f\u4e0b\u5e76\u4e0d\u5212\u7b97\u3002<\/p>\n<p>&nbsp;<\/p>\n<p><img data-dominant-color=\"c9c3c9\" data-has-transparency=\"false\" style=\"--dominant-color: #c9c3c9;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-30316\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/65319764204f2d7f78f159122d32a6e6601cc5.jpg\" width=\"340\" height=\"347\" alt=\"fig3_heatmap.png\" title=\"fig3_heatmap.png\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/65319764204f2d7f78f159122d32a6e6601cc5.jpg 340w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/65319764204f2d7f78f159122d32a6e6601cc5-294x300.jpg 294w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/65319764204f2d7f78f159122d32a6e6601cc5-147x150.jpg 147w\" sizes=\"auto, (max-width: 340px) 100vw, 340px\" \/><span class=\"data-image-title\">fig3_heatmap.png<\/span><\/p>\n<p>&nbsp;<\/p>\n<p>Figure 3 \u7ed9\u51fa\u589e\u52a0\u7ed3\u6784\u5e26\u6765\u7684\u6548\u7528\u53d8\u5316\uff0c\u7ea2\u6846\u6807\u6ce8\u7684\u4e24\u884c\u5f62\u6210\u5bf9\u7167\u3002\u4e0a\u65b9\u7b97\u672f\u4efb\u52a1 GSM8K \u5c06\u667a\u80fd\u4f53\u6570\u91cf\u4ece 1 \u589e\u81f3 4 \u4ee5\u4e0a\uff0c\u51c6\u786e\u7387\u65e0\u63d0\u5347\uff0ctoken \u6d88\u8017\u589e\u81f3\u7ea6 19 \u500d\uff0c\u6548\u7528\u4e3a\u8d1f\u3002\u4e0b\u65b9\u8d22\u52a1\u6587\u6863\u4efb\u52a1 OfficeQA \u540c\u6837\u589e\u52a0\u7ed3\u6784\uff0c\u51c6\u786e\u7387\u63d0\u5347\u7ea6 50 \u4e2a\u767e\u5206\u70b9\uff0c\u6548\u7528\u663e\u8457\u4e3a\u6b63\u3002\u540c\u4e00\u52a8\u4f5c\u5728\u4e0d\u540c\u4efb\u52a1\u4e0a\u5206\u522b\u8868\u73b0\u4e3a\u6d6a\u8d39\u4e0e\u521a\u9700\uff0c\u6784\u6210\u9010\u67e5\u8be2\u6784\u5efa\u7684\u76f4\u63a5\u52a8\u673a\u3002<\/p>\n<h4>\u56db\u3001\u5b9e\u9a8c\u7ed3\u679c<\/h4>\n<p>&nbsp;<\/p>\n<p><img data-dominant-color=\"eae9e9\" data-has-transparency=\"false\" style=\"--dominant-color: #eae9e9;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-30317\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/e217e497903a568fd34294160c1697c0ea4057.jpg\" width=\"900\" height=\"493\" alt=\"fig4_table.png\" title=\"fig4_table.png\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/e217e497903a568fd34294160c1697c0ea4057.jpg 900w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/e217e497903a568fd34294160c1697c0ea4057-300x164.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/e217e497903a568fd34294160c1697c0ea4057-150x82.jpg 150w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2026\/10\/e217e497903a568fd34294160c1697c0ea4057-768x421.jpg 768w\" sizes=\"auto, (max-width: 900px) 100vw, 900px\" \/><span class=\"data-image-title\">fig4_table.png<\/span><\/p>\n<p>&nbsp;<\/p>\n<p>\u5b9e\u9a8c\u8986\u76d6\u516d\u4e2a\u57fa\u51c6\u3001\u4e5d\u5343\u4e00\u767e\u4e5d\u5341\u4e09\u4e2a\u4efb\u52a1\uff0c\u7edf\u4e00\u91c7\u7528 Gemini 3.5 Flash \u4f5c\u4e3a\u6267\u884c\u5668\uff0c\u4e0e\u5341\u4e03\u4e2a\u57fa\u7ebf\u5bf9\u6bd4\u3002SHIFT-search \u5e73\u5747\u51c6\u786e\u7387 79.9%\uff0c\u8f83\u6700\u5f3a\u57fa\u7ebf Trace \u9ad8 7.2 \u4e2a\u767e\u5206\u70b9\uff0c\u914d\u5bf9\u81ea\u52a9\u6cd5 95% \u7f6e\u4fe1\u533a\u95f4\u4e3a 4.42% \u81f3 9.88%\u3002\u589e\u76ca\u96c6\u4e2d\u4e8e\u4e24\u4e2a\u957f\u7a0b\u3001\u5de5\u5177\u5bc6\u96c6\u578b\u57fa\u51c6\uff0cOfficeQA \u4e0a\u6240\u6709\u57fa\u7ebf\u5747\u4f4e\u4e8e 30%\uff0cSHIFT-search \u8fbe 59.3%\uff0cGAIA \u4e0a\u8fbe 68.6%\uff0c\u8f83 Trace \u5206\u522b\u9ad8 30 \u4e0e 15 \u4e2a\u767e\u5206\u70b9\u3002\u5728\u7b97\u672f\u4e0e\u591a\u8df3\u95ee\u7b54\u7b49\u77ed\u7a0b\u4efb\u52a1\u4e0a\uff0c\u5355\u4e2a agent \u5df2\u63a5\u8fd1\u4e0a\u9650\uff0c\u74f6\u9888\u4f4d\u4e8e\u6a21\u578b\u800c\u975e\u7ed3\u6784\uff0cSHIFT-search \u4e0e\u9886\u5148\u65b9\u6cd5\u57fa\u672c\u6301\u5e73\u3002<\/p>\n<p>SHIFT-value \u4e3a\u4f4e\u6210\u672c\u6863\u4f4d\uff0c\u4f9d\u636e\u9884\u6d4b\u6548\u7528\u9009\u62e9\u7ed3\u6784\uff0c\u5e73\u5747\u51c6\u786e\u7387 74.5%\uff0c\u4ecd\u9ad8\u4e8e\u5168\u90e8\u57fa\u7ebf\uff0c\u6267\u884c token \u8f83\u6700\u5f3a\u57fa\u7ebf\u51cf\u5c11 32%\u3002\u540c\u4e00\u67b6\u6784\u5e08\u53ef\u540c\u65f6\u63d0\u4f9b\u6700\u51c6\u786e\u4e0e\u6700\u7ecf\u6d4e\u7684\u4e24\u6863\u914d\u7f6e\uff0c\u6309\u9884\u7b97\u9009\u62e9\u3002<\/p>\n<h4>\u4e94\u3001\u6d88\u878d\u4e0e\u5206\u6790<\/h4>\n<p>\u6d88\u878d\u5b9e\u9a8c\u8fdb\u4e00\u6b65\u9a8c\u8bc1\u4e86\u82e5\u5e72\u8bbe\u8ba1\u9009\u62e9\u3002\u9650\u5236\u52a8\u4f5c\u7a7a\u95f4\u540e\uff0c\u4ec5\u5141\u8bb8\u52a0\u5de5\u5177\u6216\u4ec5\u5141\u8bb8\u6539\u6307\u4ee4\uff0c\u51c6\u786e\u7387\u5747\u4e0b\u964d\uff0c\u8054\u5408\u9009\u62e9\u8f83\u5355\u4e00\u7ef4\u5ea6\u6700\u591a\u9ad8 9.1 \u4e2a\u767e\u5206\u70b9\u3002\u641c\u7d22\u4e0e\u8d2a\u5fc3\u9010\u6b65\u6784\u5efa\u7684\u5e73\u5747\u51c6\u786e\u7387\u57fa\u672c\u4e00\u81f4\uff0c79.9 \u5bf9 79.8\uff0c\u4f46\u6267\u884c token \u51cf\u5c11\u7ea6\u4e09\u6210\u3002\u4ec5\u505a\u4e00\u6b65\u524d\u77bb\u7684\u4ef7\u503c\u8d2a\u5fc3\u6210\u672c\u6700\u4f4e\uff0c\u51c6\u786e\u7387\u5374\u964d\u81f3 60.2\uff0c\u539f\u56e0\u5728\u4e8e\u65e0\u6cd5\u4e3a\u9700\u8981\u540e\u7eed\u6b65\u9aa4\u624d\u663e\u73b0\u6536\u76ca\u7684\u667a\u80fd\u4f53\u8ba1\u5206\u3002\u4ef7\u503c\u5934\u7684\u9009\u62e9\u80fd\u529b\u4e5f\u7ecf\u8fc7\u5355\u72ec\u9a8c\u8bc1\uff0c\u5728\u540c\u4e00\u5019\u9009\u6c60\u4e2d\u9009\u53d6\u9884\u6d4b\u6548\u7528\u6700\u9ad8\u8005\uff0cHotpotQA \u4e0a\u8fbe 49.6\uff0c\u968f\u673a\u9009\u62e9\u4ec5 39.1\u3002\u8fc1\u79fb\u65b9\u9762\uff0c\u4ee5 GSM8K \u8bad\u7ec3\u7684\u67b6\u6784\u5e08\u4e0d\u91cd\u8bad\u76f4\u63a5\u7528\u4e8e\u66f4\u96be\u7684 MATH-500\uff0c\u7ed3\u5408\u5b66\u4e60\u5230\u7684 policy \u4e0e value \u65f6\u8868\u73b0\u6700\u597d\u3002<\/p>\n<h4>\u516d\u3001\u7ed3\u8bba\u4e0e\u5c40\u9650<\/h4>\n<p>SHIFT \u5c06\u591a\u667a\u80fd\u4f53\u7cfb\u7edf\u7684\u6784\u5efa\u4ece\u4f9d\u8d56\u7ecf\u9a8c\u4e0e\u8bd5\u9519\uff0c\u63a8\u8fdb\u5230\u53ef\u5b66\u4e60\u3001\u53ef\u9884\u6d4b\u7684\u6846\u67b6\u3002\u5b83\u4ee5\u51bb\u7ed3\u9aa8\u5e72\u3001\u4ec5\u8bad\u7ec3 LoRA \u7684\u8f7b\u91cf Gemma \u4f5c\u4e3a\u67b6\u6784\u5e08\uff0c\u628a\u8bbe\u8ba1\u6210\u672c\u4ece\u6bcf\u6b21\u63a8\u7406\u8f6c\u79fb\u5230\u4e00\u6b21\u6027\u8bad\u7ec3\u5e76\u644a\u8584\uff0c\u4e0d\u4f9d\u8d56\u66f4\u5927\u7684\u6267\u884c\u5668\uff0c\u5728\u516d\u4e2a\u57fa\u51c6\u4e0a\u540c\u65f6\u53d6\u5f97\u6700\u9ad8\u5e73\u5747\u51c6\u786e\u7387\u4e0e\u663e\u8457\u66f4\u4f4e\u7684\u6210\u672c\u6863\u4f4d\uff0c\u53ef\u76f4\u63a5\u7528\u4e8e\u667a\u80fd\u4f53\u4ea7\u54c1\u3002<\/p>\n<p>\u5176\u8fb9\u754c\u540c\u6837\u660e\u786e\u3002\u65b9\u6cd5\u8282\u7701\u7684\u662f\u6267\u884c\u5668\u8c03\u7528\u6210\u672c\uff0c\u67b6\u6784\u5e08\u8bad\u7ec3\u4e0e\u641c\u7d22\u5f00\u9500\u5355\u72ec\u8ba1\uff0c\u4ec5\u901a\u8fc7\u644a\u8584\u6d88\u5316\u3002\u5728\u6267\u884c\u5668\u80fd\u529b\u8f83\u5f3a\u3001\u7ed3\u6784\u5bf9\u6210\u8d25\u5f71\u54cd\u8f83\u5c0f\u7684\u77ed\u7a0b\u4efb\u52a1\u4e0a\uff0c\u4f18\u52bf\u6536\u7a84\uff0c\u4f5c\u8005\u4ea6\u6307\u51fa\u6b64\u7c7b\u4efb\u52a1\u7684\u74f6\u9888\u4f4d\u4e8e\u6a21\u578b\u3002\u7ed3\u6784\u9009\u62e9\u4f9d\u8d56\u9884\u6d4b\u6a21\u578b\u7684\u6548\u7528\u4f30\u8ba1\uff0c\u9884\u6d4b\u7cbe\u5ea6\u662f\u5173\u952e\uff0c\u5f53\u524d\u7ed3\u679c\u663e\u793a\u5176\u5728\u7559\u51fa\u5019\u9009\u4e2d\u80fd\u7a33\u5b9a\u9009\u51fa\u66f4\u4f18\u7ed3\u6784\u3002\u65b9\u6cd5\u80fd\u5426\u63a8\u5e7f\u5230\u66f4\u5f00\u653e\u7684\u771f\u5b9e\u5de5\u4f5c\u6d41\uff0c\u4ecd\u53d6\u51b3\u4e8e\u540e\u7eed\u590d\u73b0\u4e0e\u751f\u4ea7\u73af\u5883\u9a8c\u8bc1\u3002<\/p>\n<h4>\u8bba\u6587\u4fe1\u606f<\/h4>\n<div>\n<div class=\"hljs-cto\">\n<div class=\"hljs-cto\"><button class=\"copy_btn disable\" data-clipboard-target=\"#code_id_0\">\u590d\u5236<\/button><\/p>\n<div class=\"code-toolbar\">\n<pre class=\"has-pre-numbering language-plain\" tabindex=\"0\"><code class=\"language-plain\">\u6807\u9898    Dynamic Harness Search: Building Multi-Agent Systems Per-Query via Prediction\r\n\u4f5c\u8005    Som Sagar, Shasha Li, Hejie Cui, Ransalu Senanayake, Sercan \u00d6. Ar\u0131k\r\n\u673a\u6784    Google, Arizona State University\r\nPDF     https:\/\/arxiv.org\/pdf\/2610.04137<\/code>\r\n\r\n\u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_30309\" class=\"pvc_stats total_only  \" data-element-id=\"30309\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/www.ixpower.cn\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>SHIFT \u5c06\u591a\u667a\u80fd\u4f53\u7cfb\u7edf\u7684\u6784\u5efa\u4ece\u4f9d\u8d56\u7ecf\u9a8c\u4e0e\u8bd5\u9519\uff0c\u63a8\u8fdb\u5230\u53ef\u5b66\u4e60\u3001\u53ef\u9884\u6d4b\u7684\u6846\u67b6\u3002\u5b83\u4ee5\u51bb\u7ed3\u9aa8\u5e72\u3001\u4ec5\u8bad\u7ec3 LoRA  [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_30309\" class=\"pvc_stats total_only  \" data-element-id=\"30309\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/www.ixpower.cn\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"publish_to_discourse":"","publish_post_category":"","wpdc_auto_publish_overridden":"","wpdc_topic_tags":"","wpdc_pin_topic":"","wpdc_pin_until":"","discourse_post_id":"","discourse_permalink":"","wpdc_publishing_response":"","wpdc_publishing_error":"","footnotes":""},"categories":[3,23,20,80],"tags":[2312],"class_list":["post-30309","post","type-post","status-publish","format-standard","hentry","category-ai","category-23","category-20","category-80","tag-googleagent"],"_links":{"self":[{"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/posts\/30309","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/comments?post=30309"}],"version-history":[{"count":3,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/posts\/30309\/revisions"}],"predecessor-version":[{"id":30318,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/posts\/30309\/revisions\/30318"}],"wp:attachment":[{"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/media?parent=30309"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/categories?post=30309"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ixpower.cn\/index.php\/wp-json\/wp\/v2\/tags?post=30309"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}