{"id":12372,"date":"2026-09-18T16:48:49","date_gmt":"2026-09-18T08:48:49","guid":{"rendered":"https:\/\/infernews.com\/blog\/scienceide-turning-scientific-codebase-into-agent\/"},"modified":"2026-09-18T16:48:49","modified_gmt":"2026-09-18T08:48:49","slug":"scienceide-turning-scientific-codebase-into-agent","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/scienceide-turning-scientific-codebase-into-agent\/","title":{"rendered":"ScienceIDE \u628a\u79d1\u5b78\u6a21\u64ec\u8b8a\u6210 AI agent"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/09\/pasted-56cbba81bfa6.jpg\" alt=\"ide overview\"><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u5927\u591a\u6578 code agent \u8a13\u7df4\u74b0\u5883\u7528\u975c\u614b\u55ae\u5143\u6e2c\u8a66\u6216\u5408\u6210\u984c\u76ee\uff0cScienceIDE \u8d70\u4e00\u689d\u66f4\u72e0\u5605\u8def\uff1a\u76f4\u63a5\u642c\u771f\u5be6\u79d1\u5b78\u6a21\u64ec\u5668\u5165\u569f\u505a\u8a66\u7149\u5834\uff0c\u7b49 agent \u6539\u52d5 pinned\u3001\u672a\u7d93\u4fee\u6539\u5605\u4e0a\u6e38\u6e90\u78bc\uff0c\u518d\u7531 verifier \u91cd\u65b0\u7de8\u8b6f\u4e26\u8dd1\u539f\u672c\u5605\u7269\u7406 case\uff0c\u6578\u503c\u8981\u5c0d\u5f97\u8fd4\u5148\u7b97\u904e\u3002\u63db\u53e5\u8a71\uff0creward \u5514\u4fc2\u300cdiff \u4f3c\u5514\u4f3c\u300d\uff0c\u800c\u4fc2\u300c\u6a21\u64ec\u7d50\u679c\u5571\u5514\u5571\u300d\u3002\u5462\u5957\u6846\u67b6\u5c0d\u5de5\u7a0b\u5718\u968a\u540c\u505a AI for Science \u7814\u7a76\u5605\u4eba\u7279\u5225\u6709\u7528\uff0c\u56e0\u70ba\u4f62\u76f4\u63a5\u91cf\u5ea6 agent \u5c0d\u7269\u7406\u8a08\u7b97\u5605\u5f71\u97ff\uff0c\u800c\u5514\u4fc2\u7d19\u4e0a\u8ac7\u5175\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u63d0\u4f9b 64 \u500b\u74b0\u5883\u540c 85 \u689d ScienceIDE-Hard \u4efb\u52d9\uff0c\u5df2\u516c\u958b 15 \u500b\uff0c\u5168\u90e8\u90fd\u9644\u5e36\u53ef\u91cf\u5ea6\u5605\u96e3\u5ea6\u6307\u6a19\uff08\u672a\u4fee\u5fa9\u5206\u6578\u3001\u7de8\u8f2f\u9ede\u6578\u91cf\u3001\u9a57\u8b49\u6210\u672c\u3001\u901a\u904e\u7387\uff09\uff0c\u907f\u514d\u4eba\u70ba\u6a19 difficulty\u3002Reward \u8a2d\u8a08\u4ea6\u5e7e\u5de7\u5999\uff1auntouched repository \u4e00\u5b9a\u651e\u96f6\u5206\uff0c\u7b49\u6a21\u578b\u5fc5\u9808\u771f\u4fc2\u505a\u5622\u5148\u6709\u5206\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5718\u968a\u540c\u6642\u91cb\u51fa PhAI-IDE-4B\u30019B\u300172B \u4e09\u500b\u6a21\u578b\uff0c\u8986\u84cb\u5514\u540c\u7b97\u529b\u9810\u7b97\u3002\u5f9e\u7d50\u679c\u7747\uff0cSFT \u5f8c PLUTO-Particles-Dust \u7531 0.0 \u8df3\u5230 0.33\uff1bQwen3.5-4B \u7d93 30 \u6b65 RL\uff0cLAPS \u7531 0.357 \u5347\u5230 0.857\uff0cMITgcm-biogeo \u7531 0.286 \u5347\u5230 0.571\u3002\u8a13\u7df4\u7528\u79d1\u5b78 code \u5514\u6b62\u63d0\u5347\u5462\u985e\u4efb\u52d9\uff0c\u4ea6\u5e36\u52d5\u4e00\u822c benchmark\uff1aCodeXGLUE defect detection +6.99pp\uff0cBBH Word Sorting +36pp\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8981\u90e8\u7f72\u5605\u4eba\u6700\u76f4\u63a5\u5605\u5165\u53e3\u4fc2 Hugging Face \u5605 model collection \u540c GitHub repo\uff1b\u60f3\u7406\u89e3 reward \u908f\u8f2f\u540c\u96e3\u5ea6\u5b9a\u7fa9\uff0cpaper \u540c\u9805\u76ee\u9801\u6709\u5b8c\u6574\u4ea4\u4ee3\u3002\u8a13\u7df4\u6578\u64da\u3001\u958b\u767c\u74b0\u5883\u540c verifier \u958b\u6e90\uff0c\u7b49\u793e\u5340\u53ef\u4ee5\u55ba\u771f\u5be6\u79d1\u5b78 codebase \u4e0a\u7e7c\u7e8c scale up\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u91cd\u9ede\u6458\u8981\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u771f\u5be6\u7269\u7406\u6a21\u64ec\u505a reward<\/strong>\uff1aagent \u6539\u52d5\u4e0a\u6e38\u6e90\u78bc\u5f8c\uff0cverifier \u91cd\u7de8\u8b6f\u4e26\u6bd4\u5c0d\u7269\u7406 case \u6578\u503c\uff0c\u5514\u9760 diff \u76f8\u4f3c\u5ea6\u3002<\/li>\n<li><strong>64 \u500b\u74b0\u5883\u300185 \u689d Hard \u4efb\u52d9<\/strong>\uff1a\u5df2\u958b\u6e90 15 \u500b\uff0c\u96e3\u5ea6\u7531\u53ef\u91cf\u5ea6\u6307\u6a19\u81ea\u52d5\u6c7a\u5b9a\uff0c\u5514\u9760\u4eba\u624b\u6a19 label\u3002<\/li>\n<li><strong>\u4e09\u500b\u6a21\u578b\u898f\u6a21<\/strong>\uff1aPhAI-IDE-4B\u30019B\u300172B \u8986\u84cb\u5514\u540c\u7b97\u529b\uff0c\u8a13\u7df4\u8a2d\u5b9a\u8207\u8a55\u4f30\u689d\u4ef6\u4e00\u81f4\u3002<\/li>\n<li><strong>\u79d1\u5b78\u4efb\u52d9\u5916\u6ea2\u5230\u901a\u7528 benchmark<\/strong>\uff1aCodeXGLUE defect detection +6.99pp\uff0cBBH Word Sorting +36pp\u3002<\/li>\n<li><strong>\u5b8c\u6574\u958b\u6e90\u93c8<\/strong>\uff1acode\u3001environments\u3001tasks \u540c\u6a21\u578b\u6b0a\u91cd\u540c\u6b65\u91cb\u51fa\uff0c\u65b9\u4fbf\u7814\u7a76\u5718\u968a\u8907\u88fd\u540c\u64f4\u5c55\u3002<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/huggingface.co\/collections\/AItonomy\/scienceide-model-series\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/aitofound\/ScienceIDE\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>ScienceIDE \u7528 64 \u500b\u79d1\u5b78\u6a21\u64ec\u74b0\u5883\u8a13\u7df4 agent \u4fee bug\u3001\u91cd\u5beb\u51fd\u5f0f\uff0c\u4e26\u91cb\u51fa 4B\u30019B\u300172B \u4e09\u500b\u6a21\u578b\u3002\u4f60\u53ef\u4ee5\u628a\u5b83\u7406\u89e3\u6210\uff1a\u7528\u771f\u5be6\u7269\u7406\u6a21\u64ec\u7684\u6578\u503c\u5c0d\u4e0d\u5c0d\uff0c\u4f86\u5224\u65b7 AI \u6539\u7a0b\u5f0f\u6539\u5f97\u5571\u5514\u5571\u3002<\/p>\n","protected":false},"author":8,"featured_media":12371,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"wpai_generated_summary":"","footnotes":""},"categories":[133,116,76,127,185,197,158,199],"tags":[],"class_list":["post-12372","post","type-post","status-publish","format-standard","hentry","category-133","category-agentic","category-76","category-127","category-qwen","category-framework","category-ide","category-dataset-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12372","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=12372"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/12372\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/12371"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=12372"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=12372"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=12372"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}