{"id":6605,"date":"2025-12-06T10:04:25","date_gmt":"2025-12-06T02:04:25","guid":{"rendered":"https:\/\/nullthought.net\/?p=6605"},"modified":"2025-12-06T10:04:28","modified_gmt":"2025-12-06T02:04:28","slug":"%e4%bb%a3%e7%a0%81%e6%99%ba%e8%83%bd-%e4%bb%a3%e7%a0%81%e5%a4%a7%e6%a8%a1%e5%9e%8b%e5%ae%9e%e8%b7%b5%e6%8c%87%e5%8d%97","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=6605","title":{"rendered":"\u4ee3\u7801\u667a\u80fd\/\u4ee3\u7801\u5927\u6a21\u578b\u5b9e\u8df5\u6307\u5357"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><a href=\"https:\/\/arxiv.org\/abs\/2511.18538\" target=\"_blank\" rel=\"noreferrer noopener\">From Code Foundation Models to Agents and Applications: A Practical Guide to Code Intelligence<\/a><\/strong>\u53ef\u4ee5\u770b\u6210\u662f\u4e00\u90e8\u201c\u4ece\u6a21\u578b\u5230\u667a\u80fd\u4f53\u518d\u5230\u843d\u5730\u5e94\u7528\u201d\u7684<strong>\u4ee3\u7801\u667a\u80fd\u5b9e\u6218\u6307\u5357<\/strong>\uff0c\u65e2\u662f\u5927\u89c4\u6a21\u4ee3\u7801\u6a21\u578b\uff08Code LLM\uff09\u7684\u5168\u666f\u7efc\u8ff0\uff0c\u53c8\u975e\u5e38\u5f3a\u8c03\u5de5\u7a0b\u5b9e\u8df5\u548c\u8bad\u7ec3\u7ec6\u8282\u3002<\/p>\n\n\n\n<p>\u6587\u7ae0\u4f5c\u8005\u4e3aJian Yang\uff08\u6768\u5065\uff0c\u5317\u4eac\u822a\u7a7a\u822a\u5929\u5927\u5b66\uff09, Xianglong Liu, Weifeng Lv, Ken Deng, Shawn Guo, Lin Jing, Yizhi Li, Shark Liu, Xianzhen Luo, Yuyu Luo, Changzai Pan, Ensheng Shi, Yingshui Tan, Renshuai Tao, Jiajun Wu, Xianjie Wu, Zhenhe Wu, Daoguang Zan, Chenchen Zhang, Wei Zhang, He Zhu, Terry Yue Zhuo, Kerui Cao, Xianfu Cheng, Jun Dong, Shengjie Fang, Zhiwei Fei, Xiangyuan Guan, Qipeng Guo, Zhiguang Han, Joseph James, Tianqi Luo, Renyuan Li, Yuhang Li, Yiming Liang, Congnan Liu, Jiaheng Liu, Qian Liu, Ruitong Liu, Tyler Loakman, Xiangxin Meng, Chuang Peng, Tianhao Peng, Jiajun Shi, Mingjie Tang, Boyang Wang, Haowen Wang, Yunli Wang, Fanglin Xu, Zihan Xu, Fei Yuan, Ge Zhang, Jiayi Zhang, Xinhao Zhang, Wangchunshu Zhou, Hualei Zhu, King Zhu, Bryan Dai, Aishan Liu, Zhoujun Li, Chenghua Lin, Tianyu Liu, Chao Peng, Kai Shen, Libo Qin, Shuangyong Song, Zizheng Zhan, Jiajun Zhang, Jie Zhang, Zhaoxiang Zhang, Bo Zheng\uff0c\u6765\u81eaBUAA-SKLCCSE, Alibaba, ByteDance, M-A-P, BJTU, OPPO, HKUST (GZ), BUPT, TeleAI, Shanghai AI Lab, Manchester, StepFun, UoS, SCU, CASIA, NJU, Kuaishou, HIT, Huawei Cloud, Tencent, Monash\/CSIRO, NTU, ZJU, BIT, Ubiquant, NUS, HNU, PKU, CSU\u3002<\/p>\n\n\n\n<p>\u4e00\u3001\u8fd9\u7bc7\u7efc\u8ff0\u7684\u52a8\u673a\u4e0e\u4e3b\u8981\u8d21\u732e<\/p>\n\n\n\n<p>\u4f5c\u8005\u5148\u56de\u987e\u4e86\u4ece\u4eba\u5de5\u5199\u4ee3\u7801\u3001IDE \u8f85\u52a9\u3001\u6846\u67b6\u9a71\u52a8\u5f00\u53d1\uff0c\u4e00\u76f4\u5230\u4eca\u5929\u7684 AI-\u8f85\u52a9\u3001AI-\u9a71\u52a8\u518d\u5230\u672a\u6765 AI-\u81ea\u4e3b\u7f16\u7a0b\u7684\u6f14\u8fdb\u5386\u7a0b\uff0c\u8ba4\u4e3a\u5982\u4eca\u5df2\u7ecf\u8fdb\u5165\u201c\u4ee3\u7801\u667a\u80fd\u65f6\u4ee3\u201d\u3002\u50cf Copilot\u3001Cursor\u3001Claude Code\u3001Trae \u8fd9\u7c7b\u5de5\u5177\u8bf4\u660e\uff1a\u628a\u81ea\u7136\u8bed\u8a00\u76f4\u63a5\u7ffb\u8bd1\u6210\u53ef\u8fd0\u884c\u4ee3\u7801\u5df2\u7ecf\u4ece\u5b9e\u9a8c\u5ba4\u6982\u5ff5\u53d8\u6210\u4e86\u65e5\u5e38\u751f\u4ea7\u529b\u5de5\u5177\u3002<\/p>\n\n\n\n<p>\u4f46\u540c\u65f6\u5b58\u5728\u51e0\u4e2a\u660e\u663e\u7684\u201c\u9e3f\u6c9f\u201d\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u7814\u7a76 vs \u5b9e\u8df5<\/strong>\uff1a\u5b66\u672f\u754c\u4e3b\u8981\u5728 HumanEval\u3001APPS \u7b49\u5c0f\u51fd\u6570\u57fa\u51c6\u4e0a\u8ffd\u5206\uff0c\u800c\u5de5\u4e1a\u754c\u9762\u5bf9\u7684\u662f\u5b89\u5168\u6027\u3001\u8de8\u6587\u4ef6\u4f9d\u8d56\u3001\u5927\u4ed3\u5e93\u4e0a\u4e0b\u6587\u3001CI\/CD \u6d41\u7a0b\u96c6\u6210\u7b49\u590d\u6742\u95ee\u9898\u3002<\/li>\n\n\n\n<li><strong>\u6cdb\u5316 vs \u4e13\u7cbe<\/strong>\uff1a\u901a\u7528 LLM \u5728\u4ee3\u7801\u4e0a\u60ca\u8273\uff0c\u4f46\u5728\u4e25\u683c API \u7ea6\u675f\u3001\u5b89\u5168\u7b56\u7565\u3001\u8d85\u5927\u5de5\u7a0b\u9879\u76ee\u4e0a\uff0c\u8868\u73b0\u4ecd\u4e0d\u7a33\u5b9a\u3002<\/li>\n\n\n\n<li><strong>\u6a21\u578b\u80fd\u529b vs \u8bad\u7ec3\u914d\u65b9\u7f3a\u4e4f\u7cfb\u7edf\u603b\u7ed3<\/strong>\uff1a\u5927\u5bb6\u90fd\u5728\u5806\u6570\u636e\u3001\u5806\u7b97\u529b\u3001\u8c03 RL\uff0c\u4f46\u7f3a\u5c11\u201c\u4ece\u6570\u636e\u5230\u9884\u8bad\u7ec3\u3001SFT\u3001RL\u3001Agent\u3001\u8bc4\u6d4b\u3001\u5b89\u5168\u201d\u7684\u5b8c\u6574\u5de5\u7a0b\u7ecf\u9a8c\u3002<\/li>\n<\/ol>\n\n\n\n<p>\u56e0\u6b64\uff0c\u8bba\u6587\u7684\u76ee\u6807\u662f\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u63d0\u51fa\u4e00\u4e2a<strong>\u7edf\u4e00\u7684\u4ee3\u7801 LLM \u5206\u7c7b\u548c\u6280\u672f\u6846\u67b6<\/strong>\uff1b<\/li>\n\n\n\n<li>\u7cfb\u7edf\u68b3\u7406\u4ece\u6570\u636e\u3001\u6a21\u578b\u5230\u5bf9\u9f50\u3001RL \u7684\u5b8c\u6574\u6280\u672f\u94fe\u8def\uff1b<\/li>\n\n\n\n<li>\u603b\u7ed3\u4ee3\u7801\u4efb\u52a1\u548c\u57fa\u51c6\u4f53\u7cfb\uff0c\u5206\u6790\u5b83\u4eec\u7684\u4f18\u7f3a\u70b9\uff1b<\/li>\n\n\n\n<li>\u4e13\u95e8\u7528\u4e00\u5927\u7ae0\u8bb2\u8f6f\u4ef6\u5de5\u7a0b\u667a\u80fd\u4f53\uff08SWE Agents\uff09\uff1b<\/li>\n\n\n\n<li>\u4e13\u95e8\u7528\u4e00\u5927\u7ae0\u5efa\u6784<strong>\u4ee3\u7801\u5b89\u5168<\/strong>\u4e0e\u7ea2\u961f\u8bc4\u4f30\u6846\u67b6\uff1b<\/li>\n\n\n\n<li>\u6700\u540e\u7ed9\u51fa<strong>\u8bad\u7ec3\u914d\u65b9\u4e0e\u5b9e\u8df5\u6307\u5357<\/strong>\uff0c\u5305\u62ec\u89c4\u6a21\u5b9a\u5f8b\u3001\u8d85\u53c2\u6570\u3001\u6846\u67b6\u9009\u578b\u4e0e\u5b9e\u9a8c\u7ed3\u679c\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4e8c\u3001\u4ee3\u7801\u57fa\u7840\u6a21\u578b\uff1a\u4ece\u901a\u7528 LLM \u5230\u4e13\u7528 Code LLM<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u901a\u7528 LLM \u7684\u57fa\u7840\u4e0e\u5c40\u9650<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u8bba\u6587\u5148\u4ece Transformer \u8d77\u6b65\uff0c\u68b3\u7406\u4e86\u5bc6\u96c6\u6a21\u578b\u3001MoE\u3001\u590d\u53d1\/\u72b6\u6001\u7a7a\u95f4\u6a21\u578b\uff08\u5982 RWKV\u3001Mamba\uff09\u3001\u6269\u6563\u5f0f\u8bed\u8a00\u6a21\u578b\u4ee5\u53ca\u6df7\u5408\u67b6\u6784\u7684\u53d1\u5c55\uff0c\u8bf4\u660e\u8fd9\u4e9b\u67b6\u6784\u5982\u4f55\u652f\u6491\u4eca\u5929\u5f3a\u5927\u7684\u8bed\u8a00\u4e0e\u4ee3\u7801\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u63a5\u7740\u91cd\u70b9\u70b9\u51fa\u901a\u7528 LLM \u5728\u4ee3\u7801\u4e0a\u7684\u51e0\u4e2a\u4e0d\u8db3\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4e13\u4e1a\u5ea6\u4e0d\u591f\uff1a\u5bb9\u6613\u751f\u6210\u201c\u770b\u8d77\u6765\u5bf9\u3001\u5b9e\u9645\u4e0d\u5bf9\u201d\u7684\u4ee3\u7801\uff0c\u7ec6\u8282\u4e0a\u4e0d\u6ee1\u8db3 API \u5408\u7ea6\u3001\u5b89\u5168\u7b56\u7565\u6216\u9879\u76ee\u7ea6\u675f\u3002<\/li>\n\n\n\n<li>\u5b89\u5168\u6027\u8584\u5f31\uff1a\u5927\u89c4\u6a21\u8bc4\u4f30\u663e\u793a\uff0c\u5927\u7ea6\u4e00\u534a\u5de6\u53f3\u7684\u81ea\u52a8\u751f\u6210\u4ee3\u7801\u542b\u6709\u5df2\u77e5\u5b89\u5168\u6f0f\u6d1e\uff0c\u5373\u4fbf\u901a\u8fc7\u4e86\u529f\u80fd\u6d4b\u8bd5\u3002<\/li>\n\n\n\n<li>\u4ed3\u5e93\u7ea7\u7406\u89e3\u6709\u9650\uff1a\u957f\u4e0a\u4e0b\u6587\u4e0b\uff0c\u5173\u952e\u4fe1\u606f\u4e0d\u5728\u5f00\u5934\/\u7ed3\u5c3e\u65f6\u6027\u80fd\u660e\u663e\u4e0b\u964d\uff0c\u8de8\u6587\u4ef6\u4f9d\u8d56\u548c\u5168\u5c40\u4e0d\u53d8\u91cf\u7ecf\u5e38\u88ab\u5ffd\u7565\u3002<\/li>\n\n\n\n<li>\u591a\u6a21\u6001\/Agent \u573a\u666f\u4e0b\u7684\u5de5\u5177\u4f7f\u7528\u5bb9\u6613\u201c\u5e7b\u89c9\u201d\uff1a\u9009\u9519\u5de5\u5177\u3001\u9519\u8bef\u8c03\u7528\u65f6\u673a\u3001\u865a\u6784\u5de5\u5177\u8f93\u51fa\u7b49\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u8fd9\u4e9b\u95ee\u9898\u76f4\u63a5\u63a8\u52a8\u4e86\u4e13\u95e8\u9762\u5411\u4ee3\u7801\u7684 LLM \u53d1\u5c55\u3002<\/p>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u4ee3\u7801\u4e13\u7528\u5927\u6a21\u578b\uff08Code LLM\uff09\u8c31\u7cfb<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u8bba\u6587\u6309\u95ed\u6e90\/\u5f00\u6e90\u68b3\u7406\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u95ed\u6e90\u65b9\u9762\uff1a\n<ul class=\"wp-block-list\">\n<li><strong>GPT \u7cfb\u5217<\/strong>\u4ece GPT-3 \u5230 Codex\uff0c\u518d\u5230 GPT-4\u3001o \u7cfb\u5217\u4ee5\u53ca GPT-5 \/ GPT-5-Codex\uff0c\u80fd\u529b\u4ece\u51fd\u6570\u7ea7\u751f\u6210\u4e00\u8def\u6269\u5c55\u5230\u4ed3\u5e93\u7ea7\u4fee\u590d\u4e0e Agent \u5f0f\u5f00\u53d1\uff0c\u5728 SWE-Bench Verified \u7b49\u57fa\u51c6\u4e0a\u8fbe\u5230\u9886\u5148\u8868\u73b0\u3002<\/li>\n\n\n\n<li><strong>Gemini \/ PaLM \u7cfb\u5217<\/strong>\uff1a\u4ece PaLM \u7684\u5927\u89c4\u6a21\u89e3\u7801\u5668\u9884\u8bad\u7ec3\uff0c\u5230 PaLM2 \u7684\u591a\u8bed\u8a00\u4e0e UL2 \u53bb\u566a\uff0c\u518d\u5230 Gemini 1.x\/2.x \u7684\u539f\u751f\u591a\u6a21\u6001\u3001\u7a00\u758f MoE \u4e0e\u767e\u4e07\u7ea7\u4e0a\u4e0b\u6587\uff0c\u5f3a\u8c03\u201c\u4ee3\u7801 + \u5de5\u5177\u4f7f\u7528 + \u957f\u4e0a\u4e0b\u6587\u63a8\u7406\u201d\u7684\u7edf\u4e00\u3002<\/li>\n\n\n\n<li><strong>Claude \u7cfb\u5217<\/strong>\uff1a\u4e0e\u4ee3\u7801\u76f8\u5173\u7684\u4eae\u70b9\u662f\u957f\u4e0a\u4e0b\u6587\u3001\u591a\u6a21\u6001\u3001\u51fd\u6570\u8c03\u7528\uff0c\u4ee5\u53ca\u5e26\u7ec8\u7aef\u3001\u7f16\u8f91\u5668\u3001\u5305\u7ba1\u7406\u5668\u3001\u6d4f\u89c8\u5668\u7684\u201c\u7535\u8111\u4f7f\u7528\u6808\u201d\uff0c\u7279\u522b\u5728\u4ed3\u5e93\u7ea7\u4fee\u590d\u548c Agent \u5b89\u5168\u7b56\u7565\u4e0a\u6709\u8be6\u7ec6\u7cfb\u7edf\u5361\u63cf\u8ff0\u3002<\/li>\n\n\n\n<li>\u5176\u4ed6\u5982 Grok\u3001PanGu-Coder\u3001AlphaCode \/ AlphaCode2 \u7b49\uff0c\u5206\u522b\u5728\u957f\u4e0a\u4e0b\u6587\u3001\u591a\u8bed\u79cd\u3001\u7ade\u8d5b\u7f16\u7a0b\u548c\u8fdb\u5316\u5f0f\u4ee3\u7801\u641c\u7d22\u4e0a\u63d0\u51fa\u5404\u81ea\u7684\u65b9\u6848\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>\u5f00\u6e90\u65b9\u9762\uff1a\n<ul class=\"wp-block-list\">\n<li>\u4ee5 StarCoder\u3001Code LLaMA\u3001DeepSeek-Coder\u3001Qwen2.5-Coder \/ Qwen3-Coder\u3001IBM Granite-Code \u7b49\u4e3a\u4ee3\u8868\uff0c\u8986\u76d6\u4ece\u51e0\u4ebf\u5230\u6570\u767e\u4ebf\u4e43\u81f3 MoE \u6570\u767e B \u89c4\u6a21\uff0c\u652f\u6301\u51e0\u5341\u79cd\u7f16\u7a0b\u8bed\u8a00\u3001\u5341\u51e0\u4e07\u4ee5\u4e0a\u4e0a\u4e0b\u6587\uff0c\u5e76\u666e\u904d\u652f\u6301 FIM\uff08\u4e2d\u95f4\u586b\u5145\uff09\u7b49\u4ee3\u7801\u7279\u6709\u4efb\u52a1\u3002<\/li>\n\n\n\n<li>\u8bba\u6587\u5217\u4e86\u4e00\u5927\u8868\uff0c\u4ece\u5c42\u6570\u3001\u9690\u85cf\u7ef4\u5ea6\u3001\u4e2d\u95f4\u5c42\u5927\u5c0f\u3001\u6ce8\u610f\u529b\u7c7b\u578b\uff08GQA\/MQA\/MLA\uff09\u3001\u4e0a\u4e0b\u6587\u957f\u5ea6\u3001MoE \u6fc0\u6d3b\u53c2\u6570\u7b49\u7ef4\u5ea6\u5bf9\u6bd4\u5404\u5bb6\u6a21\u578b\uff0c\u65b9\u4fbf\u5de5\u7a0b\u9009\u578b\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n\n\n\n<ol start=\"3\" class=\"wp-block-list\">\n<li><strong>\u9884\u8bad\u7ec3\u6570\u636e\u4e0e\u672a\u6765\u8d8b\u52bf<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u8bba\u6587\u5bf9\u5f00\u6e90\u4ee3\u7801\u9884\u8bad\u7ec3\u6570\u636e\u8fdb\u884c\u4e86\u7cfb\u7edf\u68b3\u7406\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GitHub \u5927\u89c4\u6a21\u6293\u53d6\u4e0e\u6e05\u6d17\uff1b<\/li>\n\n\n\n<li>StarCoderData \u9488\u5bf9\u8bb8\u53ef\u8bc1\u3001\u8bed\u8a00\u8986\u76d6\u548c\u8d28\u91cf\u505a\u4e86\u7cbe\u7ec6\u8bbe\u8ba1\uff1b<\/li>\n\n\n\n<li>The Pile\u3001RedPajama \u7b49\u901a\u7528\u8bed\u6599\u4e2d\u7684\u4ee3\u7801\u90e8\u5206\uff1b<\/li>\n\n\n\n<li>\u8bed\u8a00\u7279\u5b9a\u6570\u636e\u96c6\u5982 CodeParrot \u53ea\u805a\u7126 Python\uff0c\u5e76\u901a\u8fc7\u5927\u89c4\u6a21\u53bb\u91cd\u66b4\u9732\u51fa\u516c\u5171\u4ee3\u7801\u4e2d\u60ca\u4eba\u7684\u91cd\u590d\u7387\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4f5c\u8005\u5f3a\u8c03\uff0c\u6570\u636e\u5de5\u7a0b\u4ece\u201c\u591a\u6293\u70b9\u4ee3\u7801\u201d\u53d1\u5c55\u4e3a\u201c\u5e26\u8bb8\u53ef\u8bc1\u8fc7\u6ee4\u3001\u53bb\u91cd\u3001\u9690\u79c1\u8131\u654f\u548c\u8d28\u91cf\u6807\u6ce8\u7684\u4e00\u6574\u5957\u900f\u660e\u7ba1\u7ebf\u201d\uff0c\u8fd9\u662f\u73b0\u4ee3\u5f00\u6e90 Code LLM \u80fd\u591f\u590d\u73b0\u4e0e\u8fed\u4ee3\u7684\u57fa\u7840\u3002<\/p>\n\n\n\n<p>\u5728\u672a\u6765\u8d8b\u52bf\u4e0a\uff0c\u8fd9\u4e00\u7ae0\u63d0\u70bc\u4e86\u4e09\u4e2a\u65b9\u5411\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ece\u901a\u7528 LLM \u5411\u4e13\u95e8 Code LLM \u6df1\u5ea6\u5206\u5316\uff0c\u5728\u4ed3\u5e93\u7ea7\u4efb\u52a1\u3001\u590d\u6742\u8c03\u8bd5\u548c\u591a\u6b65\u5f00\u53d1\u6d41\u7a0b\u4e0a\u5f62\u6210\u78be\u538b\u4f18\u52bf\uff1b<\/li>\n\n\n\n<li>\u4ece\u9759\u6001\u6587\u672c\u751f\u6210\u8d70\u5411<strong>Agent \u5316\u8bad\u7ec3<\/strong>\uff1a\u901a\u8fc7\u6267\u884c\u53cd\u9988\u3001\u8bfe\u7a0b\u5b66\u4e60\u3001\u5de5\u5177\u4f7f\u7528\u7b49\u673a\u5236\uff0c\u8ba9\u6a21\u578b\u80fd\u5728\u590d\u6742\u3001\u591a\u6b65\u7684\u8f6f\u4ef6\u573a\u666f\u4e2d\u81ea\u4e3b\u63a2\u7d22\uff1b<\/li>\n\n\n\n<li>\u7528\u89c4\u6a21\u5b9a\u5f8b\u6307\u5bfc\u201c\u79d1\u5b66\u5316\u8bad\u7ec3\u201d\uff0c\u9488\u5bf9\u4ee3\u7801\u4efb\u52a1\u7cfb\u7edf\u7814\u7a76\u53c2\u6570\u89c4\u6a21\u3001\u6570\u636e\u91cf\u3001\u7b97\u529b\u4e0e\u6027\u80fd\u7684\u5173\u7cfb\uff0c\u800c\u4e0d\u518d\u662f\u7eaf\u7ecf\u9a8c\u5806\u6599\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4e09\u3001\u4ee3\u7801\u4efb\u52a1\u3001\u57fa\u51c6\u4e0e\u8bc4\u6d4b\u4f53\u7cfb<\/p>\n\n\n\n<p>\u8bba\u6587\u7b2c\u4e09\u7ae0\u5bf9\u201c\u6211\u4eec\u5230\u5e95\u62ff\u4ec0\u4e48\u6765\u8bc4\u4f30\u4ee3\u7801 LLM\u201d\u505a\u4e86\u4e00\u4e2a\u5168\u76d8\u68b3\u7406\uff0c\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u51fd\u6570\u7ea7\u4efb\u52a1\uff1a\u8865\u5168\u3001\u751f\u6210\u3001\u4fee\u590d\u3001\u7ffb\u8bd1\u3001\u6548\u7387\u4f18\u5316\u3001\u4ee3\u7801\u95ee\u7b54\u3001\u6d4b\u8bd5\u751f\u6210\u7b49\uff1b<\/li>\n\n\n\n<li>\u4ed3\u5e93\u7ea7\u4efb\u52a1\uff1a\u591a\u6587\u4ef6\u751f\u6210\u4e0e\u8865\u5168\u3001\u9886\u57df\u4e13\u7528\u590d\u6742\u751f\u6210\uff08\u5982 SQL\u3001\u914d\u7f6e\uff09\u3001\u91cd\u6784\u4e0e Agent \u534f\u4f5c\u3001\u63d0\u4ea4\u8bf4\u660e\u751f\u6210\u3001\u8f6f\u4ef6\u5de5\u7a0b\u5168\u6d41\u7a0b\u4efb\u52a1\uff1b<\/li>\n\n\n\n<li>Agent \u4efb\u52a1\uff1a\u5de5\u5177\u4f7f\u7528\u3001\u6df1\u5ea6\u68c0\u7d22\u3001\u7f51\u9875\u641c\u7d22\u3001GUI \u64cd\u4f5c\u3001\u7ec8\u7aef\u64cd\u4f5c\u7b49\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u8bc4\u6d4b\u6307\u6807\u4e0a\uff0c\u4f5c\u8005\u533a\u5206\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4f20\u7edf\u6307\u6807\uff08\u5982\u57fa\u4e8e\u6d4b\u8bd5\u7528\u4f8b\u7684\u901a\u8fc7\u7387\uff09\uff0c\u8fdb\u4e00\u6b65\u884d\u751f\u5404\u79cd\u6269\u5c55\u4e0e\u7edf\u8ba1\u5206\u6790\uff1b<\/li>\n\n\n\n<li><strong>LLM-as-a-Judge<\/strong>\uff1a\u7528\u66f4\u5f3a\u7684\u6a21\u578b\u6765\u6253\u5206\u4ee3\u7801\u8d28\u91cf\uff0c\u4f46\u8981\u6ce8\u610f\u4e00\u81f4\u6027\u4e0e\u504f\u5dee\uff1b<\/li>\n\n\n\n<li>\u6267\u884c\u5f0f\u6307\u6807\uff1a\u901a\u8fc7\u8fd0\u884c\u4ee3\u7801\u3001\u5206\u6790\u8fd0\u884c\u8f68\u8ff9\u6765\u8bc4\u4f30\u6b63\u786e\u6027\u4e0e\u9c81\u68d2\u6027\uff1b<\/li>\n\n\n\n<li>\u591a\u667a\u80fd\u4f53\u4e0e\u9ad8\u7ea7\u63a8\u7406\u6846\u67b6\uff1a\u901a\u8fc7\u591a\u8f6e\u3001\u81ea\u53cd\u601d\u3001\u534f\u4f5c\u6765\u8861\u91cf\u6a21\u578b\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\uff1b<\/li>\n\n\n\n<li>\u4e00\u4e9b\u66f4\u72ec\u7279\u7684\u8303\u5f0f\uff0c\u4f8b\u5982\u5728\u4ea4\u4e92\u5f0f\u73af\u5883\u4e2d\u8bc4\u4f30 agent \u7684\u6210\u529f\u7387\u3001\u7a33\u5b9a\u6027\u548c\u5de5\u5177\u4f7f\u7528\u53ef\u4fe1\u5ea6\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u603b\u4f53\u89c2\u70b9\u662f\uff1a<strong>\u201c\u53ea\u770b\u662f\u5426\u901a\u8fc7\u5355\u6d4b\u8fdc\u8fdc\u4e0d\u591f\u201d<\/strong>\uff0c\u9700\u8981\u540c\u65f6\u5173\u6ce8\u53ef\u7ef4\u62a4\u6027\u3001\u5b89\u5168\u6027\u3001\u6548\u7387\u548c\u957f\u671f\u6f14\u5316\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u56db\u3001\u5bf9\u9f50\uff08Alignment\uff09\uff1aSFT\u3001\u591a\u8bed\u8a00\/\u591a\u6a21\u6001\u4e0e\u5f3a\u5316\u5b66\u4e60<\/p>\n\n\n\n<p>\u8fd9\u4e00\u5927\u7ae0\u805a\u7126\u4e8e\u201c\u5982\u4f55\u628a\u4e00\u4e2a\u57fa\u7840 Code LLM \u8c03\u6559\u6210\u771f\u6b63\u597d\u7528\u3001\u5b89\u5168\u7684\u5f00\u53d1\u52a9\u624b\u201d\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u76d1\u7763<a href=\"https:\/\/nullthought.net\/?p=6045\" target=\"_blank\" rel=\"noreferrer noopener\">\u5fae\u8c03<\/a>\uff08SFT\uff09<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u5206\u4e3a\u5355\u8f6e\u3001\u591a\u8f6e\u3001\u4ed3\u5e93\u7ea7 SFT\uff0c\u4ee5\u53ca\u5e26\u63a8\u7406\u8fc7\u7a0b\u7684 SFT\u3002\u8bba\u6587\u7ed9\u51fa\u6784\u9020\u63a8\u7406\u6570\u636e\u7684\u51b7\u542f\u52a8\u6d41\u7a0b\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6570\u636e\u6765\u6e90\uff1a\u7ade\u8d5b\u9898\u3001OJ\u3001\u771f\u5b9e\u4ed3\u5e93 issue \u7b49\uff1b<\/li>\n\n\n\n<li>\u6e05\u6d17\u4e0e\u53bb\u6c61\u67d3\uff1a\u53bb\u6389\u548c\u8bc4\u6d4b\u96c6\u5408\u91cd\u5408\u7684\u9898\u76ee\uff0c\u8fc7\u6ee4\u4f4e\u8d28\u91cf\u6570\u636e\uff1b<\/li>\n\n\n\n<li>\u9898\u76ee\u7b5b\u9009\u4e0e\u96be\u5ea6\u8bc4\u4f30\uff1b<\/li>\n\n\n\n<li>\u751f\u6210\u201c\u601d\u8def + \u89e3\u7b54\u201d\u7684\u63a8\u7406\u94fe\uff1b<\/li>\n\n\n\n<li>\u8fc7\u6ee4\u4e0e\u4fee\u6b63\uff0c\u6700\u7ec8\u6784\u9020\u9ad8\u8d28\u91cf\u7684\u63a8\u7406-\u4ee3\u7801\u6570\u636e\u96c6\u3002<\/li>\n<\/ul>\n\n\n\n<ol start=\"2\" class=\"wp-block-list\">\n<li><strong>\u591a\u8bed\u8a00\u3001\u591a\u6a21\u6001\u4ee3\u7801\u7406\u89e3<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u6587\u7ae0\u7cfb\u7edf\u56de\u987e\u4e86\u591a\u8bed\u8a00 Code LLM\uff08\u591a\u79cd\u7f16\u7a0b\u8bed\u8a00\u3001\u591a\u81ea\u7136\u8bed\u8a00\u6307\u4ee4\uff09\u4ee5\u53ca\u201c\u56fe\u50cf + \u4ee3\u7801\u201d\u7684\u591a\u6a21\u6001\u6a21\u578b\uff0c\u6bd4\u5982\u770b UI \u622a\u56fe\u81ea\u52a8\u751f\u6210\u524d\u7aef\u4ee3\u7801\u3001\u770b\u8bbe\u8ba1\u7a3f\u751f\u6210\u7f51\u9875\u3001\u89e3\u6790\u8f6f\u4ef6\u5de5\u7a0b\u6587\u6863\u548c UML \u56fe\u7b49\u3002\u4f5c\u8005\u628a\u8fd9\u770b\u4f5c\u4ee3\u7801 LLM \u771f\u6b63\u8d70\u8fdb\u201c\u771f\u5b9e\u5f00\u53d1\u573a\u666f\u201d\u7684\u5173\u952e\u80fd\u529b\u3002<\/p>\n\n\n\n<ol start=\"3\" class=\"wp-block-list\">\n<li><strong>\u5f3a\u5316\u5b66\u4e60\u4e0e\u53ef\u9a8c\u8bc1\u5956\u52b1\uff08RL &amp; RLVR\uff09<\/strong><\/li>\n<\/ol>\n\n\n\n<p>\u5728\u4ee3\u7801\u9886\u57df\uff0c\u5f3a\u5316\u5b66\u4e60\u7684\u4e00\u4e2a\u5929\u7136\u4f18\u52bf\u662f\u6709\u5927\u91cf\u201c\u53ef\u81ea\u52a8\u5224\u5206\u201d\u7684\u4fe1\u53f7\u2014\u2014\u7f16\u8bd1\u662f\u5426\u901a\u8fc7\u3001\u6d4b\u8bd5\u662f\u5426\u901a\u8fc7\u3001\u6027\u80fd\u662f\u5426\u6539\u5584\u7b49\u3002\u8bba\u6587\u68b3\u7406\u4e86\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5404\u7c7b RL \u7b97\u6cd5\u5728\u4ee3\u7801\u751f\u6210\u3001\u7406\u89e3\u3001\u8f6f\u4ef6\u5de5\u7a0b\u4efb\u52a1\u4e2d\u7684\u5e94\u7528\uff1b<\/li>\n\n\n\n<li><strong>RL with Verifiable Rewards\uff08RLVR\uff09<\/strong>\uff1a\u5229\u7528\u5355\u6d4b\u3001\u6267\u884c\u7ed3\u679c\u7b49\u53ef\u9a8c\u8bc1\u4fe1\u53f7\u6765\u5b9a\u4e49\u5956\u52b1\uff0c\u907f\u514d\u7eaf\u4e3b\u89c2\u504f\u597d\uff1b<\/li>\n\n\n\n<li><strong>\u5956\u52b1\u5851\u5f62<\/strong>\uff1a\u4e0d\u4ec5\u5956\u52b1\u201c\u662f\u5426\u901a\u8fc7\u201d\uff0c\u8fd8\u5bf9\u4ee3\u7801\u98ce\u683c\u3001\u53ef\u8bfb\u6027\u3001\u6548\u7387\u7b49\u8bbe\u7f6e\u66f4\u7cbe\u7ec6\u7684\u5956\u52b1\uff0c\u7ed3\u5408\u5b89\u5168\u7ea6\u675f\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4f5c\u8005\u5f3a\u8c03\uff0cRLVR \u6b63\u5728\u6210\u4e3a\u4ee3\u7801\u540e\u8bad\u7ec3\u7684\u6838\u5fc3\u6280\u672f\u4e4b\u4e00\uff0c\u56e0\u4e3a\u5b83\u628a\u201c\u771f\u5b9e\u4e16\u754c\u7684\u8fd0\u884c\u7ed3\u679c\u201d\u76f4\u63a5\u5f15\u5165\u8bad\u7ec3\u95ed\u73af\u3002<\/p>\n\n\n\n<p>\u4e94\u3001\u8f6f\u4ef6\u5de5\u7a0b\u667a\u80fd\u4f53\uff08SWE Agents\uff09<\/p>\n\n\n\n<p>\u8fd9\u90e8\u5206\u662f\u6574\u7bc7\u8bba\u6587\u7684\u4e00\u4e2a\u4eae\u70b9\uff0c\u4e13\u95e8\u628a\u8f6f\u4ef6\u5de5\u7a0b\u62c6\u6210\u9700\u6c42\u3001\u5f00\u53d1\u3001\u6d4b\u8bd5\u3001\u7ef4\u62a4\u3001\u7aef\u5230\u7aef\u4e94\u4e2a\u9636\u6bb5\uff0c\u5206\u6790\u5404\u9636\u6bb5\u7684 Agent \u5f62\u6001\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u9700\u6c42\u5de5\u7a0b<\/strong>\uff1a\u4ece\u9700\u6c42\u6587\u672c\u3001\u7528\u6237\u5bf9\u8bdd\u4e2d\u62bd\u53d6\u89c4\u683c\uff0c\u81ea\u52a8\u751f\u6210\u7528\u4f8b\u3001\u7528\u6237\u6545\u4e8b\u6216\u521d\u59cb\u8bbe\u8ba1\u6587\u6863\u3002<\/li>\n\n\n\n<li><strong>\u8f6f\u4ef6\u5f00\u53d1<\/strong>\uff1a\u4ece\u751f\u6210\u51fd\u6570\u3001\u6a21\u5757\uff0c\u5230\u6309 issue \u4fee\u6539\u4ee3\u7801\u3001\u8de8\u6587\u4ef6 refactor\uff0c\u518d\u5230\u4ed3\u5e93\u7ea7\u201c\u8bfb-\u6539-\u6d4b-\u63d0 PR\u201d\u7684\u5168\u6d41\u7a0b\u3002<\/li>\n\n\n\n<li><strong>\u8f6f\u4ef6\u6d4b\u8bd5<\/strong>\uff1a\u81ea\u52a8\u751f\u6210\u5355\u5143\u6d4b\u8bd5\/\u96c6\u6210\u6d4b\u8bd5\uff0c\u8bbe\u8ba1\u8986\u76d6\u66f4\u9ad8\u7684\u6d4b\u8bd5\u96c6\uff0c\u6267\u884c\u56de\u5f52\u6d4b\u8bd5\u5e76\u5206\u6790\u5931\u8d25\u539f\u56e0\u3002<\/li>\n\n\n\n<li><strong>\u8f6f\u4ef6\u7ef4\u62a4<\/strong>\uff1a\u5b9a\u4f4d bug\u3001\u4fee\u590d\u9057\u7559\u95ee\u9898\u3001\u5904\u7406\u4f9d\u8d56\u5347\u7ea7\u3001\u9605\u8bfb\u4e0e\u66f4\u65b0\u65e7\u4ee3\u7801\u3002<\/li>\n\n\n\n<li><strong>\u7aef\u5230\u7aef\u8f6f\u4ef6 Agent<\/strong>\uff1a\u5982 ChatDev\u3001MetaGPT\u3001AgileCoder\uff0c\u5c06\u4e00\u4e2a\u8f6f\u4ef6\u9879\u76ee\u4ece\u9700\u6c42\u3001\u8bbe\u8ba1\u3001\u7f16\u7801\u3001\u6d4b\u8bd5\u5230\u90e8\u7f72\u4ee5\u201c\u591a\u667a\u80fd\u4f53\u6d41\u6c34\u7ebf\u201d\u7684\u65b9\u5f0f\u4e32\u8d77\u6765\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u5728\u8bad\u7ec3\u6280\u5de7\u4e0a\uff0c\u8bba\u6587\u63d0\u51fa\u4e00\u4e2a\u5f88\u91cd\u8981\u7684\u601d\u8def\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7528 RL \u6216\u5927\u89c4\u6a21\u4ea4\u4e92\u5728\u73af\u5883\u4e2d\u201c\u63a2\u7d22\u51fa\u597d\u7b56\u7565\u201d\uff08\u4f8b\u5982 SWE-Exp \u7684\u4fee\u590d\u8f68\u8ff9\u3001\u81ea\u8fdb\u5316\u7684 SE-Agent \u63a8\u7406\u8def\u5f84\uff09\uff1b<\/li>\n\n\n\n<li>\u518d\u628a\u8fd9\u4e9b\u6210\u529f\u6848\u4f8b\u4e0e\u5931\u8d25\u6848\u4f8b\u4e2d\u7684\u7ecf\u9a8c<strong>\u56de\u704c\u5230 SFT \u6570\u636e\u4e2d<\/strong>\uff0c\u7528\u76d1\u7763\u5b66\u4e60\u5de9\u56fa\uff1b<\/li>\n\n\n\n<li>\u5f62\u6210 <strong>\u201cRL \u63a2\u7d22 + SFT \u5de9\u56fa\u201d\u7684\u81ea\u8fdb\u5316\u95ed\u73af<\/strong>\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u672a\u6765\u8d8b\u52bf\u90e8\u5206\uff0c\u5219\u628a\u7814\u7a76\u65b9\u5411\u6307\u5411\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u4ece\u5206\u6563\u5c0f Agent \u8d70\u5411\u8986\u76d6\u6574\u4e2a\u751f\u547d\u5468\u671f\u7684<strong>\u5b8c\u6574\u7f16\u6392\u7cfb\u7edf<\/strong>\uff1b<\/li>\n\n\n\n<li>\u901a\u8fc7\u77e5\u8bc6\u56fe\u8c31\u548c\u9879\u76ee\u957f\u671f\u8bb0\u5fc6\uff0c\u6784\u5efa\u5bf9\u5927\u4ed3\u5e93\u7684\u201c\u5fc3\u667a\u6a21\u578b\u201d\uff0c\u7ed3\u5408\u7248\u672c\u5386\u53f2\u3001\u8fd0\u884c\u65e5\u5fd7\u7b49\u4fe1\u606f\uff1b<\/li>\n\n\n\n<li>\u4ece\u5355\u667a\u80fd\u4f53\u8f6c\u5411\u591a\u667a\u80fd\u4f53\u751f\u6001\uff0c\u5f3a\u8c03\u89d2\u8272\u5206\u5de5\u548c\u534f\u4f5c\u3001\u81ea\u8fdb\u5316\u4e0e\u81ea\u6211\u6539\u8fdb\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u516d\u3001\u4ee3\u7801\u5728\u901a\u7528\u667a\u80fd\u4f53\u4e2d\u7684\u4e09\u79cd\u89d2\u8272<\/p>\n\n\n\n<p>\u8bba\u6587\u8fd8\u4e13\u95e8\u63d0\u51fa\u201cCode for Generalist Agents\u201d\u7684\u89c6\u89d2\uff0c\u628a\u4ee3\u7801\u770b\u6210\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u4ea4\u4e92\u534f\u8bae<\/strong>\uff1a\u901a\u8fc7\u5de5\u5177\u8c03\u7528\uff08Tool Use\uff09\u3001\u6a21\u578b\u4e0a\u4e0b\u6587\u534f\u8bae\uff08MCP\uff09\u3001\u591a\u667a\u80fd\u4f53\u534f\u4f5c\u534f\u8bae\uff0c\u8ba9\u5404\u5de5\u5177\u3001\u670d\u52a1\u3001Agent \u53ef\u4ee5\u7528\u4ee3\u7801\u6216\u7c7b\u4f3c DSL \u7684\u5f62\u5f0f\u8fdb\u884c\u6807\u51c6\u5316\u4ea4\u4e92\u3002<\/li>\n\n\n\n<li><strong>\u80fd\u529b\u672c\u8eab<\/strong>\uff1a\u667a\u80fd\u4f53\u53ef\u4ee5\u201c\u7528\u4ee3\u7801\u601d\u8003\u201d\uff08\u5199\u5c0f\u811a\u672c\u6d4b\u8bd5\u5047\u8bbe\uff09\u548c\u201c\u7528\u4ee3\u7801\u884c\u52a8\u201d\uff08\u8c03\u7528\u7ec8\u7aef\u3001API\u3001\u4eff\u771f\u73af\u5883\uff09\uff0c\u4ece\u800c\u5927\u5e45\u589e\u5f3a\u53ef\u9a8c\u8bc1\u6027\u548c\u53ef\u7ec4\u5408\u6027\u3002<\/li>\n\n\n\n<li><strong>\u73af\u5883\u63a5\u53e3<\/strong>\uff1a\u5728\u4eff\u771f\u73af\u5883\u6216\u201c\u7535\u8111\u4f7f\u7528\u201d\u573a\u666f\u4e2d\uff0c\u4ee3\u7801\u76f4\u63a5\u5c31\u662f\u73af\u5883\u64cd\u4f5c\u63a5\u53e3\uff0c\u667a\u80fd\u4f53\u901a\u8fc7\u5199\u811a\u672c\u3001\u6267\u884c\u547d\u4ee4\u6765\u64cd\u63a7\u7cfb\u7edf\u3002<\/li>\n<\/ol>\n\n\n\n<p>\u8fd9\u90e8\u5206\u7684\u6838\u5fc3\u89c2\u70b9\u662f\uff1a<strong>\u4f1a\u5199\u4ee3\u7801\u7684 Agent\uff0c\u5929\u7136\u66f4\u9002\u5408\u4f5c\u4e3a\u901a\u7528\u6570\u5b57\u4e16\u754c\u7684\u64cd\u4f5c\u7cfb\u7edf\u3002<\/strong><\/p>\n\n\n\n<p>\u4e03\u3001\u4ee3\u7801 LLM \u7684\u5b89\u5168\u95ee\u9898\u4e0e\u9632\u62a4\u6846\u67b6<\/p>\n\n\n\n<p>\u5b89\u5168\u7ae0\u8282\u9996\u5148\u6307\u51fa\u73b0\u5b9e\u95ee\u9898\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5927\u89c4\u6a21\u8bc4\u4f30\u8868\u660e\uff0c\u5373\u4fbf\u529f\u80fd\u6b63\u786e\uff0c\u5927\u91cf\u81ea\u52a8\u751f\u6210\u4ee3\u7801\u4ecd\u542b\u5df2\u77e5\u6f0f\u6d1e\uff0c\u5f00\u6e90\u6a21\u578b\u5982 Qwen3\u3001DeepSeek-R1 \u7b49\u4e5f\u96be\u4ee5\u5e78\u514d\uff1b<\/li>\n\n\n\n<li>\u9ad8\u7ea7\u653b\u51fb\u53ef\u4ee5\u8bf1\u5bfc\u6a21\u578b\u751f\u6210\u201c\u529f\u80fd\u6b63\u786e\u4f46\u6697\u85cf\u540e\u95e8\u201d\u7684\u4ee3\u7801\uff0c\u666e\u901a prompt \u5de5\u7a0b\u96be\u4ee5\u5bdf\u89c9\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u6587\u7ae0\u7ed9\u51fa\u4e86\u4e00\u4e2a\u56db\u5927\u6a21\u5757\u7684\u5b89\u5168\u6846\u67b6\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5b89\u5168\u9884\u8bad\u7ec3<\/strong>\n<ul class=\"wp-block-list\">\n<li>\u6570\u636e\u6765\u6e90\u53ef\u4fe1\u4e0e\u8bb8\u53ef\u8bc1\u5408\u89c4\uff1b<\/li>\n\n\n\n<li>\u6570\u636e\u5ba1\u8ba1\u4e0e\u6e05\u6d17\uff0c\u8fc7\u6ee4\u6076\u610f\u6a21\u5f0f\uff1b<\/li>\n\n\n\n<li>\u89c4\u907f\u9690\u79c1\u6cc4\u9732\u3001\u51cf\u5c11\u504f\u89c1\uff1b<\/li>\n\n\n\n<li>\u5bf9\u6297\u5f0f\u4ee3\u7801\u53d8\u6362\u4e0b\u4fdd\u6301\u9c81\u68d2\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u5b89\u5168\u540e\u8bad\u7ec3<\/strong>\n<ul class=\"wp-block-list\">\n<li>\u6784\u9020\u5b89\u5168\u76f8\u5173\u8bad\u7ec3\u6570\u636e\u96c6\uff08\u5305\u542b\u5b89\u5168\/\u4e0d\u5b89\u5168\u793a\u4f8b\u53ca\u89e3\u91ca\uff09\uff1b<\/li>\n\n\n\n<li>\u8fdb\u884c\u4e13\u95e8\u7684\u5b89\u5168\u5411 SFT\uff1b<\/li>\n\n\n\n<li>\u7528\u504f\u597d\u4f18\u5316\u548c RL \u8fdb\u4e00\u6b65\u524a\u5f31\u201c\u672c\u5730\u7f3a\u9677\u201d\uff0c\u4f8b\u5982\u504f\u597d\u5b89\u5168\u5b9e\u73b0\uff1b<\/li>\n\n\n\n<li>\u5c06\u5b89\u5168\u6027\u7eb3\u5165 RL \u5956\u52b1\u8bbe\u8ba1\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong><a href=\"https:\/\/nullthought.net\/?p=4543\" target=\"_blank\" rel=\"noreferrer noopener\">\u7ea2\u961f\u6d4b\u8bd5\uff08Red-teaming\uff09<\/a><\/strong>\n<ul class=\"wp-block-list\">\n<li>\u63d0\u793a\u7ea7\u64cd\u63a7\uff1a\u5404\u79cd\u7ed5\u8fc7\u5b89\u5168\u7b56\u7565\u7684 prompt \uff1b<\/li>\n\n\n\n<li>\u8bed\u4e49\u4e0e\u4e0a\u4e0b\u6587\u64cd\u63a7\uff1a\u901a\u8fc7\u590d\u6742\u573a\u666f\u8bf1\u5bfc\u6a21\u578b\u8f93\u51fa\u5371\u9669\u4ee3\u7801\uff1b<\/li>\n\n\n\n<li>Agent \u5de5\u4f5c\u6d41\u653b\u51fb\uff1a\u501f\u52a9\u5de5\u5177\u8c03\u7528\u94fe\u3001\u6587\u4ef6\u7cfb\u7edf\u64cd\u4f5c\u7b49\u653e\u5927\u5371\u5bb3\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>Agent \u7cfb\u7edf\u5c42\u9762\u7684\u9632\u62a4<\/strong>\n<ul class=\"wp-block-list\">\n<li>\u5b89\u5168\u6c99\u7bb1\u548c\u9694\u79bb\u6267\u884c\u73af\u5883\uff1b<\/li>\n\n\n\n<li>\u6267\u884c\u524d\u9759\u6001\u68c0\u6d4b\u4e0e\u89c4\u5219\u9a8c\u8bc1\uff1b<\/li>\n\n\n\n<li>\u8fd0\u884c\u65f6\u7684\u76d1\u63a7\u4e0e\u610f\u56fe\u6821\u9a8c\uff0c\u5fc5\u8981\u65f6\u4e2d\u65ad\u6216\u56de\u6eda\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n\n\n\n<p>\u8bba\u6587\u5f3a\u8c03\uff0c<strong>\u5b89\u5168\u4e0d\u53ea\u662f\u6a21\u578b\u5c42\u9762\u7684\u8fc7\u6ee4\uff0c\u800c\u662f\u6a21\u578b\u3001\u6570\u636e\u3001\u5de5\u5177\u548c\u6267\u884c\u73af\u5883\u7684\u534f\u540c\u5de5\u7a0b\u95ee\u9898<\/strong>\u3002<\/p>\n\n\n\n<p>\u516b\u3001\u8bad\u7ec3\u914d\u65b9\u4e0e\u5b9e\u8df5\u6307\u5357<\/p>\n\n\n\n<p>\u4f5c\u8005\u7528\u4e00\u6574\u7ae0\u603b\u7ed3\u8bad\u7ec3\u5b9e\u8df5\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u7b80\u8981\u4ecb\u7ecd\u5206\u5e03\u5f0f\u8bad\u7ec3\u6846\u67b6\u548c\u5e76\u884c\u7b56\u7565\uff0c\u4e3a\u8d85\u5927 Code LLM \u7684\u8bad\u7ec3\u642d\u597d\u5de5\u7a0b\u652f\u67b6\uff1b<\/li>\n\n\n\n<li>\u7ed9\u51fa\u9884\u8bad\u7ec3\u9636\u6bb5\u7684\u6307\u5bfc\uff0c\u5305\u62ec\uff1a\n<ul class=\"wp-block-list\">\n<li>\u6570\u636e\u6e05\u6d17\u4e0e\u53bb\u91cd\u7684\u5fc5\u8981\u6027\uff1b<\/li>\n\n\n\n<li>\u4ee3\u7801\u4e0e\u81ea\u7136\u8bed\u8a00\u7684\u6bd4\u4f8b\u3001\u4e0d\u540c\u7f16\u7a0b\u8bed\u8a00\u7684\u914d\u6bd4\uff1b<\/li>\n\n\n\n<li>\u5b66\u4e60\u7387\u3001\u6279\u5927\u5c0f\u3001\u4e0a\u4e0b\u6587\u957f\u5ea6\u7b49\u8d85\u53c2\u7684\u5178\u578b\u8bbe\u7f6e\u601d\u8def\uff1b<\/li>\n\n\n\n<li>\u5982\u4f55\u7ed3\u5408\u89c4\u6a21\u5b9a\u5f8b\u8bbe\u8ba1\u201c\u5728\u7b97\u529b\u9884\u7b97\u4e0b\u6700\u5212\u7b97\u201d\u7684\u6a21\u578b\u89c4\u6a21\u4e0e\u8bad\u7ec3\u6b65\u6570\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li>SFT \u9636\u6bb5\u5219\u5f3a\u8c03\uff1a\u6307\u4ee4\u6570\u636e\u8d28\u91cf\u3001\u4efb\u52a1\u591a\u6837\u6027\u3001\u4ee3\u7801\/\u975e\u4ee3\u7801\u6df7\u5408\u7b56\u7565\u4ee5\u53ca\u8bad\u7ec3\u7a33\u5b9a\u6027\u63a7\u5236\u3002<\/li>\n\n\n\n<li>RL \u9636\u6bb5\u7ed9\u51fa\uff1a\u4f55\u65f6\u503c\u5f97\u4e0a RL\u3001\u5982\u4f55\u6784\u9020\u53ef\u9a8c\u8bc1\u5956\u52b1\u3001\u5982\u4f55\u517c\u987e\u6027\u80fd\u4e0e\u5b89\u5168\u3001\u5982\u4f55\u5904\u7406\u8bad\u7ec3\u4e0d\u7a33\u5b9a\u4e0e\u63a2\u7d22-\u5229\u7528\u6743\u8861\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4e5d\u3001\u5e94\u7528\u5c42\uff1a\u4ece IDE \u52a9\u624b\u5230\u4e91\u5e73\u53f0\u4e0e\u7ec8\u7aef Agent<\/p>\n\n\n\n<p>\u6700\u540e\u4e00\u7ae0\u56de\u5230\u5e94\u7528\u89c6\u89d2\uff0c\u7cfb\u7edf\u68b3\u7406 Code LLM \u5728\u5de5\u7a0b\u5b9e\u8df5\u4e2d\u7684\u51e0\u7c7b\u5178\u578b\u5f62\u6001\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>IDE \u96c6\u6210\u5f00\u53d1\u52a9\u624b<\/strong>\uff1a\u8865\u5168\u3001\u91cd\u6784\u3001\u89e3\u91ca\u3001\u751f\u6210\u6d4b\u8bd5\u4e0e\u6587\u6863\uff0c\u6df1\u5ea6\u5d4c\u5165 VS Code\u3001JetBrains \u7b49\u73af\u5883\uff1b<\/li>\n\n\n\n<li><strong>\u4e91\u539f\u751f\u7f16\u7801\u5e73\u53f0<\/strong>\uff1a\u5728\u6d4f\u89c8\u5668\u6216\u4e91 IDE \u4e2d\u63d0\u4f9b\u534f\u4f5c\u5f0f\u7f16\u7801\u3001\u81ea\u52a8\u73af\u5883\u914d\u7f6e\u4e0e\u4e00\u952e\u90e8\u7f72\uff1b<\/li>\n\n\n\n<li><strong>\u7ec8\u7aef\u578b Agent<\/strong>\uff1a\u5728\u547d\u4ee4\u884c\u4e2d\u901a\u8fc7 LLM \u63a5\u6536\u81ea\u7136\u8bed\u8a00\u6307\u4ee4\uff0c\u81ea\u52a8\u6267\u884c\u547d\u4ee4\u3001\u7f16\u8f91\u6587\u4ef6\u3001\u8fd0\u884c\u6d4b\u8bd5\uff1b<\/li>\n\n\n\n<li><strong>\u4ee3\u7801\u4fee\u590d\u4e0e\u9a8c\u8bc1\u5e73\u53f0<\/strong>\uff1a\u628a LLM \u4e0e\u9759\u6001\u5206\u6790\u3001\u5f62\u5f0f\u5316\u9a8c\u8bc1\u5de5\u5177\u7ed3\u5408\uff0c\u505a\u81ea\u52a8\u4fee\u590d\u3001PR \u5ba1\u67e5\u548c\u8d28\u91cf\u4fdd\u8bc1\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4f5c\u8005\u5e0c\u671b\u901a\u8fc7\u8fd9\u4e00\u8282\uff0c\u628a\u524d\u9762\u6240\u6709\u5173\u4e8e\u6a21\u578b\u3001\u8bad\u7ec3\u3001Agent\u3001\u5b89\u5168\u7684\u8ba8\u8bba\u843d\u5230\u201c\u600e\u4e48\u5728\u771f\u5b9e\u56e2\u961f\u91cc\u7528\u8d77\u6765\u201d\u7684\u5c42\u9762\u3002<\/p>\n\n\n\n<p>\u5341\u3001\u6574\u4f53\u8bc4\u4ef7<\/p>\n\n\n\n<p>\u7efc\u5408\u6765\u770b\uff0c\u8fd9\u7bc7\u8bba\u6587\u7684\u4ef7\u503c\u5728\u4e8e\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8986\u76d6\u5168\u9762<\/strong>\uff1a\u4ece\u57fa\u7840\u6a21\u578b\u3001\u9884\u8bad\u7ec3\u6570\u636e\u3001\u5bf9\u9f50\u3001RL\u3001\u5b89\u5168\uff0c\u5230 SWE Agent\u3001\u901a\u7528 Agent\u3001\u8bad\u7ec3\u914d\u65b9\u548c\u5b9e\u9645\u5e94\u7528\uff0c\u51e0\u4e4e\u628a\u76ee\u524d Code LLM \u7684\u5173\u952e\u8bdd\u9898\u90fd\u7cfb\u7edf\u634b\u4e86\u4e00\u904d\u3002<\/li>\n\n\n\n<li><strong>\u5de5\u7a0b\u5bfc\u5411\u5f3a<\/strong>\uff1a\u4e0d\u662f\u7b80\u5355\u7f57\u5217\u8bba\u6587\uff0c\u800c\u662f\u660e\u786e\u7ed9\u51fa\u201c\u4ece\u6570\u636e\u5230 Agent\u201d\u7684\u5de5\u7a0b\u6d41\u6c34\u7ebf\u4e0e\u5b9e\u8df5\u5efa\u8bae\u3002<\/li>\n\n\n\n<li><strong>\u5f3a\u8c03\u672a\u6765\u65b9\u5411<\/strong>\uff1a\u7279\u522b\u662f\u5728 Agent \u5316\u8bad\u7ec3\u3001\u957f\u4e0a\u4e0b\u6587\u77e5\u8bc6\u5efa\u6a21\u3001\u96c6\u6210\u5168\u751f\u547d\u5468\u671f\u7684 SWE \u751f\u6001\u548c\u5b89\u5168\u6cbb\u7406\u65b9\u9762\u7ed9\u51fa\u4e86\u6e05\u6670\u7684\u7814\u7a76\u8def\u7ebf\u56fe\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587From Code Foundation Models to Agents and Application [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","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":"","ast-disable-related-posts":"","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":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"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":""},"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":[3,8],"tags":[39,94,86,69],"class_list":["post-6605","post","type-post","status-publish","format-standard","hentry","category-it","category-tech","tag-ai","tag-agentic-ai","tag-llm","tag-engineering"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=3\" rel=\"category\">IT<\/a> <a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587From Code Foundation Models to Agents and Application&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6605","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6605"}],"version-history":[{"count":1,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6605\/revisions"}],"predecessor-version":[{"id":6606,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/6605\/revisions\/6606"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6605"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6605"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6605"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}