{"id":20745,"date":"2026-11-04T18:03:00","date_gmt":"2026-11-04T10:03:00","guid":{"rendered":"https:\/\/dguan.com.cn\/ai\/mit%e5%9b%a2%e9%98%9f%e7%94%a8gpt-6-astra%e4%bf%ae%e6%94%b9%e5%8d%83%e9%97%ae%e5%b0%8f%e6%a8%a1%e5%9e%8b0-016%e6%9d%83%e9%87%8d%ef%bc%8c%e6%95%b0%e5%ad%a6%e6%b5%8b%e8%af%95%e6%88%90%e7%bb%a9%e5%8d%87\/"},"modified":"2026-11-04T18:03:00","modified_gmt":"2026-11-04T10:03:00","slug":"mit%e5%9b%a2%e9%98%9f%e7%94%a8gpt-6-astra%e4%bf%ae%e6%94%b9%e5%8d%83%e9%97%ae%e5%b0%8f%e6%a8%a1%e5%9e%8b0-016%e6%9d%83%e9%87%8d%ef%bc%8c%e6%95%b0%e5%ad%a6%e6%b5%8b%e8%af%95%e6%88%90%e7%bb%a9%e5%8d%87","status":"publish","type":"post","link":"https:\/\/dguan.com.cn\/ai\/mit%e5%9b%a2%e9%98%9f%e7%94%a8gpt-6-astra%e4%bf%ae%e6%94%b9%e5%8d%83%e9%97%ae%e5%b0%8f%e6%a8%a1%e5%9e%8b0-016%e6%9d%83%e9%87%8d%ef%bc%8c%e6%95%b0%e5%ad%a6%e6%b5%8b%e8%af%95%e6%88%90%e7%bb%a9%e5%8d%87\/","title":{"rendered":"MIT\u56e2\u961f\u7528GPT-6 Astra\u4fee\u6539\u5343\u95ee\u5c0f\u6a21\u578b0.016%\u6743\u91cd\uff0c\u6570\u5b66\u6d4b\u8bd5\u6210\u7ee9\u5347\u81f3\u7ea6\u516d\u500d"},"content":{"rendered":"<p>\u4e00\u4e2a8\u4ebf\u53c2\u6570\u7684\u5f00\u6e90\u5c0f\u6a21\u578b\u6570\u5b66\u6210\u7ee9\u4e0d\u4f73\uff0c\u539f\u56e0\u4e0d\u662f&#8221;\u4e0d\u4f1a\u505a\u9898&#8221;\uff0c\u800c\u662f\u7b54\u9898\u65f6\u53cd\u590d\u91cd\u590d\u540c\u6837\u7684\u63a8\u7406\u6b65\u9aa4\uff0c\u76f4\u5230\u8017\u5c3d\u8f93\u51fa\u957f\u5ea6\u9650\u5236\u3002\u9ebb\u7701\u7406\u5de5\u5b66\u9662\u8ba1\u7b97\u673a\u79d1\u5b66\u4e0e\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\uff08MIT CSAIL\uff09\u7684\u7814\u7a76\u56e2\u961f\u7ed9\u51fa\u4e86\u4e00\u79cd\u975e\u5e38\u89c4\u7684\u89e3\u6cd5\uff1a\u4e0d\u8bad\u7ec3\u3001\u4e0d\u7b97\u68af\u5ea6\uff0c\u76f4\u63a5\u8ba9OpenAI\u7684\u5927\u6a21\u578bGPT-6 Astra\u8bfb\u53d6\u5e76\u4fee\u6539\u8fd9\u4e2a\u6a21\u578b\u7684\u6743\u91cd\uff0c\u628a\u95ee\u9898\u4fee\u6389\u3002<\/p>\n<p>\u8fd9\u9879\u7814\u7a76\u4e8e11\u67082\u65e5\u4ee5\u535a\u5ba2\u5f62\u5f0f\u53d1\u5e03\uff0c\u9898\u4e3a\u300aSparks of Weight-Space Programming\u300b\uff0c\u4f5c\u8005\u4e3aRikiya Takehi\u3001Jayden Teoh\u3001Alex Zhang\u548cOmar Khattab\u3002\u7814\u7a76\u56e2\u961f\u5c06\u8fd9\u5957\u65b9\u6cd5\u547d\u540d\u4e3a&#8221;\u6743\u91cd\u7a7a\u95f4\u7f16\u7a0b&#8221;\uff08Weight-Space Programming\uff09\uff0c\u6838\u5fc3\u601d\u8def\u662f\u8ba9\u901a\u7528\u5927\u8bed\u8a00\u6a21\u578b\u50cf\u7a0b\u5e8f\u5458\u9605\u8bfb\u4ee3\u7801\u4e00\u6837\uff0c\u76f4\u63a5\u9605\u8bfb\u3001\u7406\u89e3\u5e76\u6539\u5199\u53e6\u4e00\u4e2a\u795e\u7ecf\u7f51\u7edc\u7684\u6743\u91cd\uff0c\u800c\u4e0d\u7ecf\u8fc7\u4f18\u5316\u5668\u6216\u68af\u5ea6\u4e0b\u964d\u3002\u5b9e\u9a8c\u5bf9\u8c61\u662f\u963f\u91cc\u901a\u4e49\u5343\u95ee\u7684\u5f00\u6e90\u6a21\u578bQwen3.5-0.8B\u3002<\/p>\n<p>\u8bca\u65ad\u5148\u884c\uff0c\u662f\u6574\u4e2a\u5b9e\u9a8c\u7684\u5173\u952e\u4e00\u73af\u3002\u7814\u7a76\u56e2\u961f\u7ed9GPT-6 Astra\u63d0\u4f9b\u4e86\u4e00\u4e2a\u53ef\u4ee5\u8bfb\u5199\u6a21\u578b\u6743\u91cd\u7684Python\u6c99\u7bb1\u73af\u5883\uff0c\u5141\u8bb8\u5176\u4f7f\u7528\u6570\u5b66\u7ec3\u4e60\u9898\u8fdb\u884c\u6709\u9650\u6b21\u6570\u7684\u524d\u5411\u63a8\u7406\uff0c\u4f46\u660e\u786e\u7981\u6b62\u68af\u5ea6\u8bad\u7ec3\uff0c\u4e5f\u4e0d\u5141\u8bb8\u63a5\u89e6\u6700\u7ec8\u8bc4\u6d4b\u9898\u76ee\u3002Astra\u68c0\u67e5\u540e\u53d1\u73b0\uff0cQwen3.5-0.8B\u5728\u6700\u521d\u768496\u6761\u7ec3\u4e60\u56de\u7b54\u4e2d\u670985\u6761\u649e\u4e0a\u4e8632768\u4e2atoken\u7684\u8f93\u51fa\u4e0a\u9650\uff0c\u5178\u578b\u7684&#8221;\u63a8\u7406\u5faa\u73af&#8221;\u75c7\u72b6\u2014\u2014\u8fd9\u4e2a\u6bdb\u75c5\u751a\u81f3\u5728\u8be5\u6a21\u578b\u7684\u5b98\u65b9\u6a21\u578b\u5361\u4e2d\u5c31\u66fe\u88ab\u63d0\u793a\u8fc7\u3002\u968f\u540e\uff0cAstra\u5c06\u95ee\u9898\u5b9a\u4f4d\u5230\u7b2c15\u5c42\u548c\u7b2c19\u5c42\u7684\u591a\u5c42\u611f\u77e5\u673a\u7ec4\u4ef6\uff0c\u5e76\u4ece\u4e24\u5c42\u540464\u4e2a\u8f93\u51fa\u5217\u5bf9\u5e94\u7684\u6743\u91cd\u4e2d\u51cf\u53bb\u4e0e\u91cd\u590d\u884c\u4e3a\u76f8\u5173\u7684\u65b9\u5411\u5206\u91cf\uff0c\u603b\u5171\u4fee\u6539\u4e86131072\u4e2a\u6743\u91cd\uff0c\u7ea6\u5360\u6a21\u578b\u5168\u90e8\u6743\u91cd\u76840.016%\u3002<\/p>\n<p>\u4fee\u6539\u540e\u7684\u6210\u7ee9\u53d8\u5316\u76f8\u5f53\u663e\u8457\u3002\u572850\u6b21\u524d\u5411\u63a8\u7406\u7684\u9884\u7b97\u4e0b\uff0cQwen3.5-0.8B\u5728MATH-500\u6570\u5b66\u57fa\u51c6\u4e0a\u7684\u51c6\u786e\u7387\u4ece7.9%\u5347\u81f348.1%\uff0cGSM8K\u4ece13.8%\u5347\u81f367.1%\uff1b\u539f\u672c\u672a\u88ab\u8981\u6c42\u4f18\u5316\u7684IFEval\u6307\u4ee4\u9075\u5faa\u6d4b\u8bd5\u4e5f\u4ece21.0%\u5347\u81f362.3%\u3002\u7b54\u9898\u89e6\u53ca\u8f93\u51fa\u4e0a\u9650\u7684\u6bd4\u4f8b\u5219\u4ece93%\u964d\u81f333.1%\u3002\u503c\u5f97\u4e00\u63d0\u7684\u662f\uff0c\u8fd9\u4e2a\u7ed3\u679c\u5bf9\u8c03\u7528\u9884\u7b97\u76f8\u5f53\u654f\u611f\uff1a\u5728\u524d\u5411\u63a8\u7406\u6b21\u6570\u9650\u5236\u4e3a10\u6b21\u548c25\u6b21\u65f6\uff0cMATH-500\u6210\u7ee9\u5206\u522b\u53ea\u670912.3%\u548c14.1%\uff0c\u53ea\u6709\u572850\u6b21\u9884\u7b97\u4e0b\u624d\u51fa\u73b0\u8df3\u5347\uff0c\u8fd9\u610f\u5473\u7740\u65b9\u6cd5\u7684\u5b9e\u9645\u6548\u679c\u4e0e\u8bca\u65ad\u9a8c\u8bc1\u7684\u6295\u5165\u89c4\u6a21\u5bc6\u5207\u76f8\u5173\u3002<\/p>\n<p>\u540c\u6837\u503c\u5f97\u5173\u6ce8\u7684\u662f\u5bf9\u7167\u7ec4\u7684&#8221;\u5931\u8d25&#8221;\u3002\u7814\u7a76\u56e2\u961f\u628a\u540c\u6837\u7684\u65b9\u6cd5\u7528\u4e8eGemma 3 270M\u2014\u2014\u4e00\u4e2a\u539f\u672c\u5f88\u5c11\u9677\u5165\u63a8\u7406\u5faa\u73af\u7684\u5c0f\u6a21\u578b\uff0c\u7ed3\u679cMATH-500\u6210\u7ee9\u59cb\u7ec8\u65e0\u6cd5\u63d0\u5347\u4e00\u4e2a\u767e\u5206\u70b9\u4ee5\u4e0a\u3002\u7814\u7a76\u8005\u7531\u6b64\u5f97\u51fa\u7684\u7ed3\u8bba\u9887\u4e3a\u514b\u5236\uff1a\u4fee\u6539\u6743\u91cd\u80fd\u591f\u6e05\u9664\u906e\u6321\u5df2\u6709\u80fd\u529b\u7684\u969c\u788d\uff0c\u4f46\u65e0\u6cd5\u51ed\u7a7a\u6dfb\u52a0\u6a21\u578b\u539f\u672c\u4e0d\u5177\u5907\u7684\u80fd\u529b\u3002\u6362\u53e5\u8bdd\u8bf4\uff0c\u8fd9\u6b21\u5b9e\u9a8c\u5e76\u975e&#8221;\u7ed9\u5c0f\u6a21\u578b\u704c\u6ce8\u65b0\u667a\u80fd&#8221;\uff0c\u800c\u662f\u89e3\u5f00\u6346\u4f4f\u5b83\u7684\u7ef3\u5b50\u3002<\/p>\n<p>\u9664\u4fee\u590d\u5b9e\u9a8c\u5916\uff0c\u7814\u7a76\u8fd8\u5305\u62ec\u591a\u9879\u80fd\u529b\u6d4b\u8bd5\uff1aAstra\u4ec5\u51ed17253\u4e2a\u6743\u91cd\u6570\u503c\uff0c\u5c31\u5728\u6ca1\u6709\u6e38\u620f\u753b\u9762\u548c\u540d\u79f0\u9009\u9879\u7684\u60c5\u51b5\u4e0b\uff0c\u6b63\u786e\u8bc6\u522b\u51fa\u56db\u4e2a\u7b56\u7565\u7f51\u7edc\u5206\u522b\u8bad\u7ec3\u81ea\u6253\u7816\u5757\u3001\u4e52\u4e53\u7403\u3001\u5403\u8c46\u4eba\u548c\u8fc7\u9a6c\u8def\u56db\u6b3e\u6e38\u620f\uff1b\u5b83\u8fd8\u80fd\u5728\u65e0\u8bad\u7ec3\u6570\u636e\u3001\u65e0\u68af\u5ea6\u7684\u6761\u4ef6\u4e0b\u4ece\u96f6\u7f16\u5199\u7f51\u7edc\u6743\u91cd\uff0c\u751f\u6210\u7684\u6e38\u620f\u7b56\u7565\u7f51\u7edc\u5f97\u5206\u8fdc\u8d85PPO\u7b97\u6cd5\u8bad\u7ec31\u4ebf\u6b65\u7684\u57fa\u7ebf\u3002\u5f53\u7136\uff0c\u8fd9\u9879\u5de5\u4f5c\u5c1a\u672a\u7ecf\u540c\u884c\u8bc4\u5ba1\uff0c\u4e5f\u6709\u7814\u7a76\u8005\u6307\u51fa\uff0c\u5b9e\u9a8c\u63d0\u793a\u8bcd\u5df2\u63d0\u524d\u5f15\u5bfcAstra\u6392\u67e5\u91cd\u590d\u63a8\u7406\u95ee\u9898\uff0c\u5176\u81ea\u4e3b\u6027\u53ef\u80fd\u88ab\u9ad8\u4f30\u3002<\/p>\n<p>\u7814\u7a76\u56e2\u961f\u5df2\u516c\u5f00\u5b8c\u6574\u7684\u5b9e\u9a8c\u65e5\u5fd7\u548c\u6743\u91cd\u4fee\u6539\u8bb0\u5f55\uff0c\u5e76\u5c06\u7f16\u8f91\u540e\u7684\u6a21\u578b\u53d1\u5e03\u5728\u5f00\u6e90\u793e\u533a\u3002\u5bf9\u5728\u672c\u5730\u90e8\u7f72\u5c0f\u6a21\u578b\u7684\u5f00\u53d1\u8005\u800c\u8a00\uff0c\u8fd9\u63d0\u4f9b\u4e86\u4e00\u6761\u8bad\u7ec3\u4e4b\u5916\u7684\u6545\u969c\u4fee\u590d\u601d\u8def\uff1b\u800c\u5bf9\u6574\u4e2a\u884c\u4e1a\u6765\u8bf4\uff0c\u5f53\u6a21\u578b\u5f00\u59cb\u8bfb\u53d6\u548c\u4fee\u6539\u6a21\u578b\uff0c\u6a21\u578b\u5b89\u5168\u7684\u5ba1\u67e5\u89c6\u91ce\u6216\u8bb8\u4e5f\u8981\u4ece\u5355\u4e2a\u6a21\u578b\u6269\u5c55\u5230\u6a21\u578b\u4e4b\u95f4\u7684\u5173\u7cfb\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e00\u4e2a8\u4ebf\u53c2\u6570\u7684\u5f00\u6e90\u5c0f\u6a21\u578b\u6570\u5b66\u6210\u7ee9\u4e0d\u4f73\uff0c\u539f<\/p>\n","protected":false},"author":10,"featured_media":20744,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"colormag_page_container_layout":"default_layout","colormag_page_sidebar_layout":"default_layout","footnotes":""},"categories":[1],"tags":[],"class_list":["post-20745","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"acf":[],"_links":{"self":[{"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/posts\/20745","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/users\/10"}],"replies":[{"embeddable":true,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/comments?post=20745"}],"version-history":[{"count":0,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/posts\/20745\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/media\/20744"}],"wp:attachment":[{"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/media?parent=20745"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/categories?post=20745"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dguan.com.cn\/ai\/wp-json\/wp\/v2\/tags?post=20745"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}