{"id":11428,"date":"2022-12-31T18:03:00","date_gmt":"2022-12-31T10:03:00","guid":{"rendered":"http:\/\/139.9.1.231\/?p=11428"},"modified":"2023-01-02T16:53:06","modified_gmt":"2023-01-02T08:53:06","slug":"codex","status":"publish","type":"post","link":"http:\/\/139.9.1.231\/index.php\/2022\/12\/31\/codex\/","title":{"rendered":"OpenAI \u4ee3\u7801\u751f\u6210\u6a21\u578b Codex\uff1a Evaluating Large Language Models Trained on Code"},"content":{"rendered":"\n\n\n

Codex<\/h2>\n\n\n\n

https:\/\/openai.com\/blog\/openai-codex\/<\/a><\/p>\n\n\n\n

Evaluating Large Language Models Trained on Code<\/a><\/strong><\/p>\n\n\n\n

Copilot<\/a>\u7684\u6838\u5fc3\u6280\u672f\uff1a\u7ed9\u5b9a\u51fd\u6570\u540d\u548c\u529f\u80fd\u63cf\u8ff0\uff0c\u53ef\u4ee5\u81ea\u52a8\u8fdb\u884c\u4ee3\u7801\u8865\u5168<\/strong>\uff0c\u6216\u8005\u7ed9\u5b9a\u4ee3\u7801\uff0c\u7ed9\u51fa\u76f8\u5173\u6587\u6863<\/strong>\u3002\u4f5c\u8005\u56e2\u961f\u6536\u96c6\u4e86Github\u4e0a\u6240\u6709\u7684\u4e0d\u91cd\u590d\u7684python\u4ee3\u7801\uff0c\u603b\u8ba1179GB\uff0c\u5e76\u8fdb\u884c\u4e86\u7b80\u5355\u8fc7\u6ee4(\u53bb\u6389\u4e86\u8fc7\u5927\u7684\u6587\u4ef6\uff08>1MB\uff09\u548c\u8fc7\u957f\u7684\u4ee3\u7801(>100\u884c\u6216\u5355\u884c\u8d85\u8fc71000\u4e2a\u5b57\u7b26))\uff0c\u5728\u6570\u636e\u96c6\u4e0a\u9762\u8bad\u7ec3\u4e86\u4e00\u4e2aGPT3\u6a21\u578b\u3002<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

\u4f5c\u8005\u56e2\u961f\u624b\u52a8\u7f16\u5199\u4e86164\u4e2a\u51fd\u6570\uff08\u907f\u514d\u6570\u636e\u6cc4\u6f0f\uff09\uff0c\u6bcf\u4e2a\u51fd\u6570\u5305\u62ec\u4ee3\u7801\u3001\u6587\u6863\u4ee5\u53ca\u5355\u5143\u6d4b\u8bd5<\/a>\uff0c\u5e73\u5747\u6bcf\u4e2a\u95ee\u9898\u5305\u62ec7.7\u4e2a\u6d4b\u8bd5\u6837\u4f8b\uff0c\u7528\u4e8e\u8bc4\u4f30\u6a21\u578b\u3002Codex 12\u4ebf\u53c2\u6570\u7684\u6a21\u578b\u80fd\u89e3\u51b328.8%\u7684\u95ee\u9898\uff0c3\u4ebf\u53c2\u6570\u7684\u6a21\u578b\u80fd\u89e3\u51b313.2%\u7684\u95ee\u9898\uff0c\u4f5c\u8005\u56e2\u961f\u53c8\u6536\u96c6\u4e86\u4e00\u4e2a\u8ddf\u6d4b\u8bd5\u96c6\u5dee\u4e0d\u591a\u7684\u6570\u636e\u96c6\u7528\u4e8e\u6a21\u578b\u5fae\u8c03\uff0c\u5fae\u8c03\u4ee5\u540e\uff0c\u5f97\u5230Codex-S\u53ef\u4ee5\u89e3\u51b337.7%\u7684\u95ee\u9898\u3002\u800c\u4f7f\u7528 repeated sampling\uff0c\u5373\u8fd0\u884c\u4e00\u767e\u6b21\u6a21\u578b\uff0c\u53ea\u8981\u6709\u4e00\u4e2a\u8f93\u51fa\u89e3\u51b3\u4e86\u95ee\u9898\u5c31\u7b97\u6210\u529f\u7684\u8bdd\uff0c\u90a3\u4e48Codex-S\u80fd\u89e3\u51b377.5%\u7684\u95ee\u9898\uff08CodeX\u80fd\u89e3\u51b370.2%\uff09\uff0c\u800c\u5982\u679c\u9009\u62e9100\u4e2a\u8f93\u51fa\u4e2d\u6982\u7387\u6700\u9ad8\u7684\u8f93\u51fa\uff0c\u5219\u80fd\u89e3\u51b344.5%\u7684\u95ee\u9898\u3002<\/p>\n\n\n\n

\u7ec6\u8282<\/h3>\n\n\n\n

1. \u76ee\u6807\u51fd\u6570\u6ca1\u6709\u4f7f\u7528BLEU\uff08\u56f0\u60d1\u5ea6\uff09\uff0c\u56e0\u4e3a\u4ee3\u7801\u4e0d\u540c\u4e8e\u81ea\u7136\u8bed\u8a00\uff0c\u5373\u4f7f\u7279\u522b\u76f8\u4f3c\uff0c\u4f46\u4ecd\u7136\u53ef\u80fd\u4e0d\u662f\u4e00\u4e2a\u5408\u6cd5\u7684\u8bed\u53e5\uff0c\u4f5c\u8005\u4f7f\u7528\uff1a<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

\u6765\u8bc4\u4f30\u6a21\u578b\uff0c\u5373\u751f\u6210n\u4e2a\u8f93\u51fa\uff08n>k\uff09\uff0c\u4ece\u4e2d\u968f\u673a\u62bd\u53d6k\u4e2a\u8f93\u51fa\uff0c\u8fd9k\u4e2a\u8f93\u51fa\u53ea\u8981\u6709\u4e00\u4e2a\u80fd\u901a\u8fc7\u5355\u5143\u6d4b\u8bd5\u7684\u6982\u7387.<\/p>\n\n\n\n

\u4ee3\u7801\u8fd1\u4f3c\u8ba1\u7b97pass@k\uff08\u4e3a\u4ec0\u4e48\u8981\u8fd1\u4f3c\uff1a\u5982\u679ck\uff0cn\u5f88\u5927\uff0c\u8ba1\u7b97\u5f88\u590d\u6742\uff09<\/p>\n\n\n\n

\"\"<\/figure><\/div>\n\n\n\n

2. \u8f93\u51fa\u4ee3\u7801\u7684\u6d4b\u8bd5\u5728\u6c99\u76d2\u4e2d\u8fdb\u884c\uff08\u751f\u6210\u7684\u4ee3\u7801\u53ef\u80fd\u662f\u6076\u610f\u7684\uff0c\u4f1a\u8ba9\u4f60\u7684\u673a\u5668\u51fa\u73b0\u95ee\u9898\uff09<\/p>\n\n\n\n

3. \u5728GPT3\u539f\u6709\u6a21\u578b\u4e0a\u5fae\u8c03\u5e76\u4e0d\u80fd\u53d6\u5f97\u66f4\u597d\u7684\u6548\u679c\uff0c\u4f46\u4f1a\u52a0\u901f\u6536\u655b<\/p>\n\n\n\n

4\u3001\u4ee3\u7801\u91cc\u9762\u7684\u7a7a\u683c\u5982\u679c\u4e0d\u505a\u5904\u7406\u4f1a\u5e26\u6765\u5f88\u591a\u4e0d\u5fc5\u8981\u7684\u8bcd\u8fdb\u53bb\uff0c\u5bf9\u7a7a\u683c\u505a\u7279\u6b8a\u5904\u7406\u540e\u4f1a\u51cf\u5c1130%\u7684\u8bcd<\/p>\n\n\n\n

5. \u5f53\u6a21\u578b\u8f93\u51fa\u2018\\nclass\u2019, \u2018\\ndef\u2019, \u2018\\n#\u2019, \u2018\\nif\u2019, or\u2018\\nprint\u2019\u7b49\u8bed\u53e5\u65f6\uff0c\u6a21\u578b\u4f1a\u7ec8\u6b62\u63a8\u7406\uff0c\u8f93\u51fa\u7ed3\u679c<\/p>\n\n\n\n

6. \u4f7f\u7528nucleus sampling\uff08\u6838\u91c7\u6837\uff09\uff1a\u9009\u62e9\u6982\u7387\u603b\u548cp=95%\u7684\u524dk\u4e2a\u8f93\u51fa\u7528\u4e8e\u8bc4\u4f30\u6a21\u578b<\/p>\n\n\n\n

7. \u5bf9\u8f93\u51fa\u505asoftmax\u5f97\u5230\u6982\u7387\u4e4b\u524d\uff0c\u4f1a\u9664\u4ee5\u4e00\u4e2a\u8d85\u53c2\u6570Temperature\uff0c\u6765\u8c03\u8282\u4e0d\u540c\u8f93\u51fa\u4e4b\u95f4\u7684\u6982\u7387\u5dee\u8ddd\uff0c\u5f53pass@k\u4e2d\u7684\u91c7\u6837\u6570k\u8d8a\u5927\u65f6\uff0cT\u8d8a\u5927\u6548\u679c\u8d8a\u597d<\/p>\n\n\n\n

8. \u6536\u96c6\u4e86\u8ddf\u6d4b\u8bd5\u96c6\u7c7b\u4f3c\u7684\u6570\u636e\u96c6\u7528\u4e8e\u5fae\u8c03\uff0c1\uff09\u4ece\u5404\u79cd\u6bd4\u8d5b\u4e2d\u6536\u96c6\u8d5b\u9898\uff08\u5927\u7ea6\u4e00\u4e07\u4e2a\uff09\uff0c2\uff09\u4eceContinuous Integration\u4e2d\u6536\u96c6\u4e86\u7ea640000\u4e2a\u51fd\u6570\u548c\u5355\u5143\u6d4b\u8bd5\uff0c\u5e76\u8fc7\u6ee4\uff08CodeX\u5bf9\u6bcf\u4e2a\u95ee\u9898\u751f\u6210\u4e00\u767e\u4e2a\u8f93\u51fa\uff0c\u5982\u679c\u80fd\u89e3\u51b3\u901a\u8fc7\u6d4b\u8bd5\u7528\u4f8b\u5219\u4fdd\u7559\u8be5\u6837\u672c\uff0c\u53cd\u4e4b\u5219\u53bb\u6389\uff08\u4e0d\u80fd\u901a\u8fc7\u8868\u793a\u8be5\u95ee\u9898\u592a\u96be\u6216\u6d4b\u8bd5\u7528\u4f8b\u6709\u95ee\u9898\uff09\uff09\uff0c\u5728\u8fd9\u4e2a\u6570\u636e\u96c6\u4e0a\u7ee7\u7eed\u8bad\u7ec3\uff0c\u8bad\u7ec3\u65b9\u5f0f\u76f8\u540c\uff0c\u53ea\u662f\u8be5\u6570\u636e\u96c6\u6709\u201c\u6807\u51c6\u7b54\u6848\u201d\uff0c\u5f97\u5230\u6a21\u578bCode-S<\/p>\n\n\n\n

9. \u4f7f\u7528\u6536\u96c6\u5230\u7684github\u6570\u636e\u96c6\uff0c\u91cd\u65b0\u8bad\u7ec3\u4e00\u4e2aGPT3\u6a21\u578b\u7528\u4e8e\u53cd\u5411\u751f\u6210\u6587\u6863\uff0cCodex-D\uff0c\u8bc4\u6d4bCodex-D\u6a21\u578b\u597d\u574f\u7684\u65b9\u5f0f\u662f\uff0c\u4e00\u662f\u4eba\u9605\u8bfb\u6587\u6863\u8bc4\u6d4b\u6a21\u578b\u597d\u574f\uff0c\u4e8c\u662f\u4f7f\u7528\u751f\u6210\u7684\u6587\u6863\u91cd\u65b0\u751f\u6210\u4ee3\u7801\uff0c\u770b\u80fd\u5426\u901a\u8fc7\u5355\u5143\u6d4b\u8bd5<\/p>\n\n\n\n

\u6a21\u578b\u5c40\u9650\u6027<\/h3>\n\n\n\n

1. \u6837\u672c\u6709\u6548\u6027\u4e0d\u591f\uff0c\u9700\u8981\u8bad\u7ec3\u5f88\u591a\u7684\u4ee3\u7801\uff0c\u6a21\u578b\u624d\u80fd\u8f93\u51fa\u6bd4\u8f83\u7b80\u5355\u7684\u5b9e\u9a8c<\/p>\n\n\n\n

2. Prompt\u5e94\u8be5\u600e\u4e48\u5199\u624d\u80fd\u83b7\u5f97\u6bd4\u8f83\u7406\u60f3\u7684\u4ee3\u7801\uff0c\u4f5c\u8005\u627e\u4e8613 basic building block\uff08\u5bf9\u5b57\u7b26\u4e32\u505a\u4e00\u4e9b\u7b80\u5355\u7684\u64cd\u4f5c\uff1a\u5982\u6539\u53d8\u5927\u5c0f\u5199\u3001\u53d8\u6362\u4f4d\u7f6e\u7b49\uff09\uff0c\u5c06\u6587\u6863\u5757\u4efb\u610f\u4e32\u8d77\u6765\uff0c\u53d1\u73b0\u6587\u6863\u8d8a\u957f\uff0c\u751f\u6210\u4ee3\u7801\u7684\u8d28\u91cf\u8d8a\u5dee\uff0c\u8bf4\u660edocstring\u4e0d\u5b9c\u8fc7\u957f<\/p>\n\n\n\n

3. \u5bf9\u4e8e\u7cbe\u786e\u3001\u590d\u6742\u7684\u6570\u5b66\u95ee\u9898\u5f88\u96be\u751f\u6210\u6b63\u786e\u7684\u4ee3\u7801<\/p>\n\n\n\n

\u6a21\u578b\u6f5c\u5728\u7684\u5f71\u54cd<\/h3>\n\n\n\n

1. \u8fc7\u5ea6\u4f9d\u8d56\uff1a\u4eba\u53ef\u80fd\u4f1a\u8fc7\u5ea6\u4f9d\u8d56\u751f\u6210\u7684\u4ee3\u7801\uff0c\u5982\u679c\u4f7f\u7528\u8005\u4e0d\u4ed4\u7ec6\u5ba1\u67e5\u4ee3\u7801\uff0c\u53ef\u80fd\u4f1a\u7ed9\u7a0b\u5e8f\u5e26\u6765\u6f5c\u5728\u7684\u95ee\u9898<\/p>\n\n\n\n

2. Misalignment\uff1a\u6a21\u578b\u8db3\u591f\u590d\u6742\u7684\u65f6\u5019\uff0c\u53ef\u80fd\u80fd\u8f93\u51fa\u671f\u671b\u7684\u4ee3\u7801\uff0c\u4f46\u5982\u679c\u7ed9\u5b9a\u4e00\u4e2adocstring\uff0c\u53ef\u80fd\u53ea\u80fd\u8f93\u51fa\u4e00\u4e2a\u8ddf\u8bad\u7ec3\u6570\u636e\u98ce\u683c\u76f8\u4f3c\uff0c\u770b\u4e0a\u53bb\u6b63\u786e\uff0c\u4f46\u5e76\u4e0d\u662f\u671f\u671b\u7684\u4ee3\u7801<\/p>\n\n\n\n

3. github\u7537\u6027\u7528\u6237\u5c45\u591a\uff0c\u6240\u5199\u7684\u4ee3\u7801\u53ef\u80fd\u5305\u542b\u6027\u522b\u504f\u89c1<\/p>\n\n\n\n

4. \u5e02\u573a\u548c\u7ecf\u6d4e\uff1a\u5f88\u591a\u7a0b\u5e8f\u5458\u53ef\u80fd\u4f1a\u5931\u4e1a\uff1f\u5982\u679c\u8bad\u7ec3\u6570\u636e\u91cc\u7684\u4ee3\u7801\u5bf9\u4e8e\u67d0\u4e9b\u5305\u4f7f\u7528\u8f83\u591a\uff0c\u53ef\u80fd\u5bfc\u81f4\u67d0\u4e9b\u7279\u522b\u7684\u5de5\u5177\u4f7f\u7528\u7387\u589e\u591a\u3002<\/p>\n\n\n\n

5. \u5b89\u5168\uff1a\u53ef\u80fd\u67d0\u4e9b\u4eba\u7528\u5b83\u5199\u75c5\u6bd2\u548c\u6076\u610f\u8f6f\u4ef6<\/p>\n\n\n\n

6. \u8bad\u7ec3\u8fd9\u6837\u4e00\u4e2a\u6a21\u578b\u9700\u8981\u4f7f\u7528\u5f88\u591a\u8d44\u6e90<\/p>\n\n\n\n

7. \u6cd5\u5f8b\uff1a\u4f7f\u7528\u7684\u662f\u516c\u5f00\u4ee3\u7801\uff0cfair use\uff08\u5bf9\u516c\u5171\u793e\u4f1a\u6709\u597d\u5904\u7684\u8bdd\u5e76\u6ca1\u6709\u4ec0\u4e48\u95ee\u9898\uff09\uff0c\u4f46\u7528\u4e8e\u5546\u4e1a\u884c\u4e3a\u53ef\u80fd\u4f1a\u6709\u6cd5\u5f8b\u98ce\u9669\uff0c\u751f\u6210\u7684\u4ee3\u7801\u53ef\u80fd\u8ddf\u522b\u4eba\u4e00\u6a21\u4e00\u6837\uff0c\u53ef\u80fd\u5b58\u5728\u6284\u88ad\u522b\u4eba\u5177\u6709\u7248\u6743\u6216\u8005\u4e13\u5229\u4fdd\u62a4\u7684\u4ee3\u7801\u7684\u98ce\u9669\u3002<\/p>\n\n\n\n

<\/a>\u603b\u7ed3<\/h3>\n\n\n\n

\u4f5c\u8005\u722c\u4e86\u5f88\u591agithub\u7684\u4ee3\u7801\uff0c\u8bad\u7ec3\u4e86\u4e00\u4e2aGPT3\u7684\u6a21\u578b\uff0c\u4e3a\u4e86\u8bc4\u4f30\u6a21\u578b\u7684\u6548\u679c\uff0c\u51c6\u5907\u4e86146\u5230\u9898\u7528\u4e8e\u6d4b\u8bd5\uff0c\u53d1\u73b0\u5927\u6982\u80fd\u89e3\u51b3\u5927\u698230%\u7684\u9898\uff0c\u6548\u679c\u8fd8\u4e0d\u9519\uff0c\u4e3a\u4e86\u8fdb\u4e00\u6b65\u63d0\u9ad8\u5206\u6570\uff0c\u53c8\u6536\u96c6\u4e86\u4e00\u4e2a\u8ddf\u6d4b\u8bd5\u96c6\u76f8\u4f3c\u7684\u6570\u636e\u96c6\uff0c\u5728\u4e0a\u9762\u5fae\u8c03\u3002<\/p>\n\n\n\n

GitHub Copilot<\/a><\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

Copilot \u76f8\u6bd4\u8bba\u6587codex\u4e2d\u7684\u533a\u522b\uff1a\u6a21\u578b\u90fd\u662f\u91c7\u7528GPT3\uff0c\u4f46\u662f Copilot \u4f7f\u7528\u7684\u6570\u636e\u96c6\u4e0d\u4ec5\u4ec5\u662fpython\uff0c\u8fd8\u6709\u5176\u4ed6\u8bed\u8a00\u7684\u4ee3\u7801\u4f5c\u4e3a\u6570\u636e\u96c6\u3002 GitHub \u4e0a\u516c\u5f00\u53ef\u7528\u5b58\u50a8\u5e93\u7684\u6570\u5341\u4ebf\u884c\u4ee3\u7801\u7684\u8bad\u7ec3 \u3002<\/p>\n\n\n\n

Copilot \u4f5c\u4e3a\u4e00\u4e2a\u8f85\u52a9\u7f16\u7a0b\u5de5\u5177\uff0cGitHub Copilot \u53ef\u4ee5\u901a\u8fc7\u63d0\u4f9b\u81ea\u52a8\u5b8c\u6210\u6837\u5f0f\u7684\u5efa\u8bae\u6765\u5e2e\u52a9\u4f60\u7f16\u5199\u4ee3\u7801\u3002GitHub Copilot \u662f\u4e00\u4e2a AI \u914d\u5bf9\u7a0b\u5e8f\u5458\uff0c\u53ef\u5728\u7f16\u5199\u4ee3\u7801\u65f6\u63d0\u4f9b\u81ea\u52a8\u5b8c\u6210\u6837\u5f0f\u7684\u5efa\u8bae\u3002 \u53ef\u4ee5\u4ece GitHub Copilot \u63a5\u6536\u5efa\u8bae\uff0c\u65b9\u6cd5\u662f\u5f00\u59cb\u7f16\u5199\u8981\u4f7f\u7528\u7684\u4ee3\u7801\uff0c\u6216\u8005\u7f16\u5199\u63cf\u8ff0\u4ee3\u7801\u8981\u6267\u884c\u7684\u64cd\u4f5c\u7684\u81ea\u7136\u8bed\u8a00\u6ce8\u91ca\u3002 GitHub Copilot \u4f1a\u5206\u6790\u4f60\u6b63\u5728\u7f16\u8f91\u7684\u6587\u4ef6\u4ee5\u53ca\u76f8\u5173\u6587\u4ef6\u4e2d\u7684\u4e0a\u4e0b\u6587\uff0c\u5e76\u5728\u6587\u672c\u7f16\u8f91\u5668\u4e2d\u63d0\u4f9b\u5efa\u8bae\u3002 GitHub Copilot \u7531 OpenAI Codex \u63d0\u4f9b\u652f\u6301\uff0cOpenAI Codex \u662f\u4e00\u4e2a\u7531 OpenAI \u521b\u5efa\u7684\u65b0 AI \u7cfb\u7edf\u3002<\/p>\n\n\n\n

\u4e0d\u4ec5\u662f\u5173\u952e\u5b57\u7684\u81ea\u52a8\u8865\u5168\uff0c\u8bed\u6cd5\u5efa\u8bae\uff0c\u8c03\u8bd5\u5efa\u8bae\u7b49\u3002\u800c\u662f\u5e2e\u52a9\u5f00\u53d1\u8005\u66f4\u5feb\u901f\u7684\u5b8c\u6210\u4e1a\u52a1\u4ee3\u7801\u7f16\u5199\u3002\u7b80\u800c\u8a00\u4e4b\uff0cGitHub Copilot \u662f\u4e00\u79cd AI \u5de5\u5177\uff0c\u53ef\u6839\u636e\u547d\u540d\u6216\u8005\u6b63\u5728\u7f16\u8f91\u7684\u4ee3\u7801\u4e0a\u4e0b\u6587\u4e3a\u5f00\u53d1\u8005\u63d0\u4f9b\u4ee3\u7801\u5efa\u8bae\u3002<\/p>\n\n\n\n

\u6839\u636e\u5b98\u65b9\u4ecb\u7ecd\uff0cCopilot \u5df2\u7ecf\u63a5\u53d7\u4e86\u6765\u81ea GitHub \u4e0a\u516c\u5f00\u53ef\u7528\u5b58\u50a8\u5e93\u7684\u6570\u5341\u4ebf\u884c\u4ee3\u7801\u7684\u8bad\u7ec3\uff0c\u5b83\u652f\u6301\u5927\u591a\u6570\u7f16\u7a0b\u8bed\u8a00\uff0c\u4f46\u5b98\u65b9\u5efa\u8bae\u4f7f\u7528 Python\u3001JavaScript\u3001TypeScript\u3001Ruby \u548c Go\u3002Copilot \u662f GitHub \u548cOpenAI<\/a>\u5408\u4f5c\u7684\u7ed3\u679c\uff0c OpenAI\u5f97\u5230\u4e86\u5fae\u8f6f\u7684\u5927\u529b\u652f\u6301\u3002\u5b83\u7531\u4e00\u4e2a\u540d\u4e3a Codex \u7684\u5168\u65b0 AI \u7cfb\u7edf\u63d0\u4f9b\u652f\u6301\uff0c\u8be5\u7cfb\u7edf\u57fa\u4e8e GPT-3 \u6a21\u578b<\/em><\/strong>\u3002<\/p>\n\n\n\n

\u540e\u7eed\u5de5\u4f5c\uff1a<\/h2>\n\n\n\n

DeepMind AlphaCode\u00a0<\/h3>\n\n\n\n

DeepMind\u63a8\u51fa\u4e86\u81ea\u52a8\u5199\u7b97\u6cd5\u7ade\u8d5b\u9898\u7684AI AlphaCode\uff0c\u5ba3\u79f0\u76ee\u524d\u5728Codeforces\u6bd4\u8d5b\u4e2d\u80fd\u6392\u5230\u4e2d\u4f4d\u6570\u3002Transformer + \u8d85\u5927\u6570\u636e\u96c6\u6765\u505acode generation\u3002\u867d\u7136\u73b0\u5728\u4e5f\u6709\u5f88\u591a\u5de5\u4f5c\u7528transformer\u505a\u4ee3\u7801\u9884\u8bad\u7ec3\uff0c\u6216\u8005\u505a\u4ee3\u7801\u7ffb\u8bd1\u6216\u8005\u751f\u6210\u3002\u4f46\u662f\u4ece\u8fd9\u4e48\u957f\u7684\u9898\u9762\u53bb\u751f\u6210\u7ade\u8d5b\u7684\u4ee3\u7801\u786e\u5b9e\u662f\u5934\u4e00\u6b21\u3002<\/strong><\/p>\n\n\n\n

AlphaCode \u53c2\u52a0\u7684\u662f\u4e00\u4e2a\u540d\u4e3a Codeforces \u7684\u5728\u7ebf\u7f16\u7a0b\u5e73\u53f0\u3002\u867d\u7136\u6211\u5e76\u4e0d\u719f\u6089 Codeforces\uff0c\u4f46\u66fe\u7ecf\u4e3a\u4e86\u51c6\u5907\u9762\u8bd5\u5237\u8fc7 LeetCode\u3002\u5982\u679c\u8bf4 LeetCode \u5c31\u662f\u4e3a\u4e86\u7a0b\u5e8f\u5458\u8fdb\u4e92\u8054\u7f51\u5927\u5382\u5237\u9898\u800c\u751f\uff0c\u4e3b\u8981\u8003\u5bdf\u7a0b\u5e8f\u5458\u7684\u7b97\u6cd5\u548c\u6570\u636e\u7ed3\u6784\u7684\u80fd\u529b\u7684\u8bdd\uff0c\u90a3 Codeforces \u662f\u4e00\u4e2a\u7ade\u8d5b\u7248\u7684 LeetCode\uff0cCodeforces \u4e0a\u7684\u9898\u76ee\u66f4\u50cf ACM ICPC \u6216\u8005\u4fe1\u606f\u5b66\u5965\u6797\u5339\u514b\u7ade\u8d5b\u3002<\/p>\n\n\n\n

Codeforces \u4e0a\u7684\u9898\u76ee\u4e94\u82b1\u516b\u95e8\uff0c\u4f46\u662f\u90fd\u9700\u8981\u53c2\u8d5b\u8005\u7f16\u7a0b\u6c42\u89e3\u3002\u6bcf\u4e2a\u9898\u76ee\u6709\u63cf\u8ff0\uff0c\u6709\u8f93\u5165\u6837\u4f8b\uff0c\u6709\u6b63\u786e\u7684\u8f93\u51fa\u6837\u4f8b\uff0c\u5373test cases\u3002\u5982\u679c\u63d0\u4ea4\u7684\u7a0b\u5e8f\u80fd\u591f\u5c06\u6240\u6709test cases\u90fd\u8dd1\u51fa\u6b63\u786e\u7684\u7ed3\u679c\uff0c\u90a3\u4e48\u5c31\u7b97\u8be5\u9898\u901a\u8fc7\u3002\u4e00\u9053\u9898\u53ea\u670910\u6b21\u8bd5\u9519\u673a\u4f1a\u3002<\/p>\n\n\n\n

\"\"\/
\u00a0AlphaCode \u6240\u6c42\u89e3\u7684\u95ee\u9898\u6837\u4f8b\uff0c\u6df1\u8272\u7684\u4e0a\u534a\u90e8\u5206\u4e3a\u7f16\u7a0b\u95ee\u9898\u63cf\u8ff0\uff0c\u6d45\u8272\u7684\u4e0b\u534a\u90e8\u5206\u4e3a AlphaCode \u751f\u6210\u7684\u4ee3\u7801\u7b54\u6848<\/figcaption><\/figure>\n\n\n\n

Training\uff1a\u6a21\u578b\u8bad\u7ec3\u200b<\/a><\/h3>\n\n\n\n

AlphaCode \u4f7f\u7528\u7684\u7ecf\u5178\u7684\u9884\u8bad\u7ec3+\u5fae\u8c03\uff08Pretraining + Fine-tuning\uff09\u8303\u5f0f\u3002<\/p>\n\n\n\n

\u9884\u8bad\u7ec3\u4f7f\u7528\u7684\u662f\u4ece GitHub \u722c\u4e0b\u6765\u7684\u5f00\u6e90\u4ee3\u7801\uff0c\u7ecf\u8fc7\u4e86\u7cbe\u7ec6\u7684\u9884\u5904\u7406\u548c\u6e05\u6d17\uff0c\u5927\u7ea6\u6709715GB\u3002\u770b\u5230\u8fd9\u4e2a\u89c4\u6a21\u7684\u8bad\u7ec3\u6570\u636e\uff0c\u5c31\u77e5\u9053\u53ea\u6709\u5c48\u6307\u53ef\u6570\u7684\u51e0\u5bb6\u5de8\u65e0\u9738\u516c\u53f8\u80fd\u591f\u505a\u8fd9\u4e2a\u9884\u8bad\u7ec3\uff0c\u5b9e\u5728\u662f\u592a\u5927\u4e86\uff0c\u4f30\u8ba1\u9700\u8981\u6210\u5343\u4e0a\u4e07\u5757GPU\u3002\u9884\u8bad\u7ec3\u90e8\u5206\u5355\u7eaf\u5c31\u662f\u8ba9\u6a21\u578b\u5b66\u4e60\u4e0d\u540c\u7f16\u7a0b\u8bed\u8a00\u7684\u5957\u8def\uff0c\u6216\u8005\u8bf4\u5b66\u4e60\u7f16\u7a0b\u8bed\u8a00\u4e2d\u7684\u8bed\u4e49\u548c\u8bed\u6cd5\u3002<\/p>\n\n\n\n

\u5fae\u8c03\u90e8\u5206\u4f7f\u7528\u7684\u662f CodeContests \u6570\u636e\u96c6\uff0c\u8fd9\u4e2a\u6570\u636e\u96c6\u6536\u96c6\u4e86\u5f88\u591a\u7c7b\u4f3c Codeforces \u8fd9\u6837\u7684\u7f16\u7a0b\u5e73\u53f0\u4e0a\u7684\u7f16\u7a0b\u9898\u76ee\u3001\u5143\u6570\u636e\u4ee5\u53ca\u4eba\u7c7b\u6b63\u786e\u548c\u9519\u8bef\u7684\u4ee3\u7801\u63d0\u4ea4\u7ed3\u679c\u3002\u76ee\u7684\u662f\u9488\u5bf9 Codeforces \u8fd9\u6837\u7684\u7f16\u7a0b\u7ade\u8d5b\uff0c\u8ba9\u6a21\u578b\u5b66\u4f1a\u5982\u4f55\u751f\u6210\u5bf9\u5e94\u7684\u4ee3\u7801\u3002\u8fd9\u4e2a\u6570\u636e\u96c6\u5927\u7ea62GB\u3002<\/p>\n\n\n\n

AlphaCode \u4e3b\u8981\u4f7f\u7528\u4e86\u7f16\u7801\u5668-\u89e3\u7801\u5668\uff08Encoder-Decoder\uff09\u7684 seq2seq \u65b9\u5f0f\u5efa\u6a21\u3002seq2seq \u6700\u7ecf\u5178\u7684\u5e94\u7528\u662f\u673a\u5668\u7ffb\u8bd1\u3002\u7ed9\u5b9a\u6e90\u6587\u672c\u5185\u5bb9\uff0cEncoder \u5c06\u81ea\u7136\u8bed\u8a00\u7f16\u7801\u4e3a\u4e00\u4e9b\u5411\u91cf\uff0cDecoder \u6839\u636e\u5411\u91cf\u5c06\u81ea\u7136\u8bed\u8a00\u89e3\u7801\u4e3a\u76ee\u6807\u6587\u672c\u3002\u90a3\u4e48\u5bf9\u4e8eAI\u81ea\u52a8\u5199\u4ee3\u7801\u8fd9\u4e2a\u95ee\u9898\uff0c\u5c31\u662f\u8f93\u5165\u7f16\u7a0b\u9898\u76ee\uff0c\u8ba9\u6a21\u578b\u751f\u6210\u76ee\u6807\u4ee3\u7801\u3002<\/p>\n\n\n\n

Sampling & Evaluation\uff1a\u6d77\u91cf\u8bd5\u9519\u200b<\/a><\/h3>\n\n\n\n
\"\"\/
\u56fe AlphaCode\u67b6\u6784\u56fe<\/figcaption><\/figure>\n\n\n\n

\u4e0a\u56fe\u4e3a AlphaCode \u7684\u67b6\u6784\uff0c\u5de6\u4fa7\uff08Data\uff09\u4e3a\u6a21\u578b\u548c\u6570\u636e\u90e8\u5206\uff0c\u4e3b\u8981\u4f7f\u7528 Transformer \u8fdb\u884c\u9884\u8bad\u7ec3\u548c\u5fae\u8c03\uff0c\u53f3\u4fa7\uff08Samping & Evaluation\uff09\u662f\u5982\u4f55\u751f\u6210\u4ee3\u7801\u5e76\u53c2\u4e0e Codeforces \u6bd4\u8d5b\u3002<\/p>\n\n\n\n

AlphaCode \u4f7f\u7528\u4e86\u7ecf\u5178\u7684 Transformer \u6a21\u578b\u3002\u6709\u5173 Transformer \u7684\u4ecb\u7ecd\uff0c\u7f51\u7edc\u4e0a\u5df2\u7ecf\u6709\u4e0d\u5c11\uff0c\u6211\u81ea\u5df1\u4e4b\u524d\u4e5f\u5199\u8fc7\u4e00\u4e9b Transformer<\/a> \u548c BERT<\/a> \u7684\u5165\u95e8\u6587\u7ae0\u3002\u5173\u6ce8\u6df1\u5ea6\u5b66\u4e60\u7684\u670b\u53cb\u90fd\u77e5\u9053\uff0cTransfomer \u4f5c\u4e3a\u5f53\u524d\u5927\u7ea2\u5927\u7d2b\u7684AI\u6a21\u578b\uff0c\u867d\u7136\u5728\u5404\u4e2a\u699c\u5355\u4e0a\u5237\u699c\uff0c\u4f46\u5b83\u5e76\u4e0d\u5177\u6709\u4eba\u7c7b\u57fa\u672c\u7684\u63a8\u7406\u80fd\u529b\u3002<\/p>\n\n\n\n

\u76f8\u6bd4Transformer\uff0c\u6211\u8ba4\u4e3a\u4f7f\u5f97 AlphaCode \u6210\u529f\u7684\u4e3b\u8981\u5728\u4e8e\u8fd9\u4e2a Sampling & Evaluation\u3002\u8fd9\u4e2a Sampling & Evaluation \u7cfb\u7edf\u6709\u70b9\u7c7b\u4f3c\u641c\u7d22\u5f15\u64ce\u6216\u8005\u63a8\u8350\u5f15\u64ce\u3002AI\u62e5\u6709\u5b58\u50a8\u548c\u5236\u4f5c\u6d77\u91cf\u5185\u5bb9\u7684\u80fd\u529b\uff0c\u4f46\u65e0\u6cd5\u77e5\u9053\u4eba\u7c7b\u771f\u6b63\u9700\u8981\u4ec0\u4e48\u3002\u6700\u5173\u952e\u7684\u5c31\u662f\u5982\u4f55\u4ece\u6d77\u91cf\u5185\u5bb9\u4e2d\u8fdb\u884c\u7b5b\u9009\u3002\u641c\u7d22\u6216\u63a8\u8350\u5f15\u64ce\u4e00\u822c\u4f1a\u5bf9\u6d77\u91cf\u5185\u5bb9\u8fdb\u884c\u68c0\u7d22\uff0c\u6700\u7ec8\u5448\u73b0\u7ed9\u7528\u6237\u7684\u53ea\u6709\u51e0\u6761\u5185\u5bb9\u3002\u6d77\u91cf\u7684\u5185\u5bb9\u9700\u8981\u7ecf\u8fc7\u51e0\u5927\u6b65\u9aa4\uff1a\u53ec\u56de\u3001\u7c97\u6392\u3001\u7cbe\u6392\u3001\u91cd\u6392\u3002\u5176\u5b9e\u5c31\u662f\u5148\u4ece\u6d77\u91cf\u7684\u5185\u5bb9\u5e93\u4e2d\uff0c\u5148\u7c97\u7565\u7b5b\u9009\u51fa\u4e00\u4e07\u7bc7\u7684\u5185\u5bb9\uff0c\u518d\u4f7f\u7528\u66f4\u7cbe\u7ec6\u7684\u6a21\u578b\u5bf9\u4e00\u4e07\u7bc7\u8fdb\u884c\u4e00\u6b21\u6b21\u7b5b\u9009\uff0c\u6700\u7ec8\u9009\u62e9\u51fa\u4e0e\u7528\u6237\u9700\u6c42\u6700\u76f8\u5173\u7684\u51e0\u7bc7\u5185\u5bb9\u3002<\/p>\n\n\n\n

AlphaCode \u4f7f\u7528\u4e86\u4e00\u4e2a Transformer \u6a21\u578b\uff0c\u6839\u636e\u7f16\u7a0b\u9898\u76ee\u63cf\u8ff0\uff0c\u751f\u6210\u767e\u4e07\u4efd\u4ee3\u7801\uff0c\u8fd9\u4e9b\u751f\u6210\u7684\u4ee3\u7801\u4e2d99%\u53ef\u80fd\u6839\u672c\u8dd1\u4e0d\u901a\u3002AlphaCode \u4f7f\u7528\u7f16\u7a0b\u9898\u76ee\u4e2d\u7684test cases\uff0c\u9a8c\u8bc1\u8fd9\u4e9b\u751f\u6210\u7684\u4ee3\u7801\uff0c\u8fd9\u4e2a\u8fc7\u7a0b\u4f1a\u8fc7\u6ee4\u638999%\u7684\u9519\u8bef\u4ee3\u7801\u3002<\/p>\n\n\n\n

\u7ecf\u8fc7\u8fc7\u6ee4\u4e4b\u540e\uff0c\u4ecd\u7136\u53ef\u80fd\u6709\u4e0a\u5343\u4efd\u4ee3\u7801\u80fd\u8dd1\u901a\uff0c\u800c\u4e14\u8fd9\u4e9b\u80fd\u8dd1\u901a\u9898\u76ee\u7ed9\u51fa\u7684\u6d4b\u8bd5\u6837\u4f8b\u7684\u4ee3\u7801\u4e2d\u5f88\u591a\u975e\u5e38\u76f8\u4f3c\u3002\u4e00\u4e2a\u7f16\u7a0b\u9898\u76ee\u53ea\u670910\u6b21\u63d0\u4ea4\u673a\u4f1a\uff0c\u6bcf\u4e00\u6b21\u63d0\u4ea4\u7684\u673a\u4f1a\u90fd\u975e\u5e38\u73cd\u8d35\u3002\u4e0d\u53ef\u80fd\u5c06\u4e0a\u5343\u4efd\u4ee3\u7801\u90fd\u63d0\u4ea4\u4e0a\u53bb\u3002AlphaCode \u8fd9\u65f6\u5019\u505a\u4e86\u4e00\u4e2a\u805a\u7c7b\uff08Clustering\uff09\u3002\u9996\u5148\uff1aAlphaCode \u4f7f\u7528\u4e86\u7b2c\u4e8c\u4e2a Transformer \u6a21\u578b\uff0c\u6839\u636e\u7f16\u7a0b\u9898\u76ee\u4e2d\u7684\u6587\u5b57\u63cf\u8ff0\uff0c\u81ea\u52a8\u751f\u6210\u4e00\u4e9btest cases\u3002\u4f46\u662f\u751f\u6210\u7684test cases\u5e76\u4e0d\u4fdd\u8bc1\u51c6\u786e\u6027\uff0c\u5b83\u662f\u4e3a\u4e86\u63a5\u4e0b\u6765\u7684\u805a\u7c7b\u7528\u7684\u3002\u7136\u540e\uff1a\u5c06\u751f\u6210\u7684test cases\u5582\u7ed9\u90a3\u4e9b\u4ee3\u7801\uff0c\u5982\u679c\u4e00\u4e9b\u4ee3\u7801\u7684\u751f\u6210\u7ed3\u679c\u8fd1\u4e4e\u4e00\u6837\uff0c\u8bf4\u660e\u8fd9\u4e9b\u4ee3\u7801\u80cc\u540e\u7684\u7b97\u6cd5\u6216\u903b\u8f91\u76f8\u4f3c\uff0c\u53ef\u88ab\u5f52\u4e3a\u4e00\u7c7b\u3002\u6587\u7ae0\u79f0\uff0c\u7ecf\u8fc7\u805a\u7c7b\u4e4b\u540e\uff0c\u4ece\u6570\u76ee\u8f83\u5927\u7684\u7c7b\u4e2d\u9009\u51fa\u4ee3\u7801\u53bb\u63d0\u4ea4\uff0c\u66f4\u6709\u80dc\u7b97\u3002<\/p>\n\n\n\n

\"\"\/<\/figure>\n\n\n\n

\u4e0a\u56fe\u6f14\u793a\u4e86\u8fd9\u4e2a\u8fc7\u7a0b\uff0c\u5927\u81f4\u5305\u62ec\u56db\u6b65\uff1a<\/p>\n\n\n\n

  1. \u6839\u636e\u7f16\u7a0b\u9898\u76ee\u4e2d\u7684\u63cf\u8ff0\u7b49\u4fe1\u606f\uff0c\u4f7f\u7528\u7b2c\u4e00\u4e2aTransformer\u6a21\u578b\uff0c\u751f\u6210\u767e\u4e07\u4efd\u4ee3\u7801\u3002<\/li>
  2. \u4f7f\u7528\u7f16\u7a0b\u9898\u76ee\u4e2d\u7684\u6d4b\u8bd5\u6837\u4f8btest cases\u9a8c\u8bc1\u8fd9\u767e\u4e07\u4efd\u4ee3\u7801\uff0c\u628a\u4e0d\u80fd\u901a\u8fc7\u7684\u8fc7\u6ee4\u6389\uff0c\u5269\u4e0b\u5927\u7ea6\u4e0a\u5343\u4efd\u4ee3\u7801\u3002<\/li>
  3. \u4f7f\u7528\u7b2c\u4e8c\u4e2aTransformer\u6a21\u578b\uff0c\u751f\u6210\u4e00\u4e9btest cases\u3002<\/li>
  4. \u4f7f\u7528\u7b2c3\u6b65\u751f\u6210\u7684test cases\uff0c\u5bf9\u7b2c2\u6b65\u7559\u4e0b\u7684\u4ee3\u7801\u8fdb\u884c\u9a8c\u8bc1\u5e76\u805a\u7c7b\uff0c\u5982\u679c\u4e24\u4efd\u4ee3\u7801\u5f97\u5230\u7684\u7ed3\u679c\u76f8\u540c\uff0c\u5219\u5206\u5230\u540c\u4e00\u7c7b\u3002\u7ecf\u8fc7\u805a\u7c7b\u540e\uff0c\u6700\u7ec8\u7559\u4e0b10\u7c7b\u4ee3\u7801\u3002<\/li><\/ol>\n\n\n\n

    Capabilities & Limitations\uff1a\u80fd\u529b\u548c\u9650\u5236\u200b<\/a><\/h3>\n\n\n\n

    \u6df1\u5ea6\u5b66\u4e60\u662f\u9ed1\u76d2\u6a21\u578b\uff0c\u6211\u4eec\u4e0d\u77e5\u9053\u5230\u5e95\u6a21\u578b\u5b66\u5230\u4e86\u4ec0\u4e48\uff0c\u80fd\u5426\u50cf\u4eba\u7c7b\u4e00\u6837\u8ba4\u77e5\u548c\u63a8\u7406\u3002\u8bba\u6587\u82b1\u4e86\u5f88\u5927\u7cbe\u529b\u548c\u7bc7\u5e45\u8ba8\u8bba\u4e86 AlphaCode \u7684\u80fd\u529b\u548c\u9650\u5236\u3002<\/p>\n\n\n\n

    \u4f5c\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u8bba\u70b9\uff0c\u5373 AlphaCode \u5e76\u4e0d\u662f\u5355\u7eaf\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u5bfb\u627e\u76f8\u4f3c\u89e3\u6cd5\uff0c\u6216\u8005\u8bf4 AlphaCode \u5e76\u4e0d\u662f\u5355\u7eaf\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u62f7\u8d1d\u4ee3\u7801\u3002\u4f5c\u8005\u7684\u9a8c\u8bc1\u65b9\u6cd5\u662f\u5bf9\u6bd4\u4e86\u751f\u6210\u7684\u4ee3\u7801\u548c\u8bad\u7ec3\u96c6\u4e2d\u7684\u4ee3\u7801\u4e2d\u7684\u4ee3\u7801\u7247\u6bb5\u91cd\u5408\u7684\u60c5\u51b5\uff0c\u6216\u8005\u8bf4\u68c0\u9a8c AlphaCode \u662f\u4e0d\u662f\u5355\u7eaf\u4ece\u8bad\u7ec3\u96c6\u91cc\u627e\u4e00\u4e9b\u6838\u5fc3\u4ee3\u7801\u7247\u6bb5\u5e76\u76f4\u63a5\u62f7\u8d1d\u8fc7\u6765\u3002\u56e0\u6b64\uff0c\u4f5c\u8005\u4eec\u8ba4\u4e3a\uff0cAlphaCode \u5177\u6709\u89e3\u51b3\u65b0\u95ee\u9898\u7684\u80fd\u529b\uff0c\u800c\u4e0d\u662f\u7167\u732b\u753b\u864e\u5730\u628a\u8bad\u7ec3\u6570\u636e\u62f7\u8d1d\u642c\u8fd0\u8fc7\u6765\u3002\u77e5\u4e4e\u4e0a\u6709\u4fe1\u606f\u5b66\u7ade\u8d5b\u9009\u624b\u611f\u6168\uff0c\u6709\u4e9b\u9898\u76ee\u5bf9\u4e8e\u4eba\u7c7b\u4e13\u4e1a\u9009\u624b\u6765\u8bf4\u90fd\u5f88\u96be\u5feb\u901f\u60f3\u51fa\u89e3\u6cd5\uff0c\u4f46 AlphaCode \u5374\u80fd\u591f\u5f97\u5230\u7b54\u6848\u3002<\/p>\n\n\n\n

    \u4f5c\u8005\u4eec\u53d1\u73b0\uff0c\u6a21\u578b\u751f\u6210\u7684\u4ee3\u7801\u975e\u5e38\u4f9d\u8d56\u7f16\u7a0b\u9898\u76ee\u4e2d\u7684\u63cf\u8ff0\u3002\u6bd4\u5982\uff0c\u540c\u6837\u4e00\u4e2a\u89e3\u6cd5\uff0c\u9898\u76ee\u63cf\u8ff0\u8d8a\u5197\u957f\uff0cAlphaCode \u7684\u6c42\u89e3\u51c6\u786e\u5ea6\u8d8a\u4f4e\u3002\u4f46\u662f\u5bf9\u7f16\u7a0b\u9898\u76ee\u7684\u4e00\u4e9b\u5176\u4ed6\u6539\u53d8\u5bf9\u6c42\u89e3\u5f71\u54cd\u4e0d\u5927\uff0c\u6bd4\u5982\u66f4\u6539\u53d8\u91cf\u540d\u3001\u540c\u4e49\u8bcd\u66ff\u6362\u7b49\u3002<\/p>\n\n\n\n

    \u603b\u7ed3\u200b<\/a><\/h3>\n\n\n\n

    \u4f5c\u8005\u8ba4\u4e3a\uff0cAlphaCode \u80fd\u591f\u51fb\u8d25\u534a\u6570\u4eba\u7c7b\u9009\u624b\uff0c\u4e3b\u8981\u539f\u56e0\u5728\u4e8e\uff1a<\/p>\n\n\n\n

    1. \u8bad\u7ec3\u6570\u636e\u8db3\u591f\u5927\u4e14\u8d28\u91cf\u9ad8\u3002<\/li>
    2. Transformer \u9884\u8bad\u7ec3\u6a21\u578b\u80fd\u591f\u5c06\u8bad\u7ec3\u6570\u636e\u4e2d\u6db5\u76d6\u7684\u77e5\u8bc6\u7f16\u7801\u5230\u6a21\u578b\u4e2d\u3002<\/li>
    3. Sampling & Evaluation \u7684\u6d77\u91cf\u8bd5\u9519\u673a\u5236\uff0c\u5148\u751f\u6210\u6d77\u91cf\u53ef\u80fd\u7684\u7b54\u6848\uff0c\u518d\u4e00\u6b65\u6b65\u7f29\u5c0f\u641c\u7d22\u7a7a\u95f4\u3002<\/li><\/ol>\n\n\n\n

      \u9605\u8bfb\u5b8c\u8bba\u6587\u548c\u4e00\u4e9b\u89e3\u8bfb\u4e4b\u540e\uff0c\u6211\u611f\u89c9\u81f3\u5c11\u77ed\u671f\u5185\uff0c\u79bbAI\u66ff\u4ee3\u7a0b\u5e8f\u5458\u5e94\u8be5\u8fd8\u6709\u4e00\u6bb5\u8ddd\u79bb\u3002\u4f46\u662f\uff0c\u672a\u6765\uff0c\u53ef\u771f\u4e0d\u597d\u8bf4…<\/p>\n\n\n\n

      \"\"
      \u4e0d\u5b8c\u5168\u7edf\u8ba1\u7684Code\u9884\u8bad\u7ec3\u6a21\u578b<\/figcaption><\/figure>\n\n\n\n

      \u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684CodeXGLEU\uff0c\u662f\u8fd1\u51e0\u5e74\u5bf9\u4ee3\u7801\u667a\u80fd\u4efb\u52a1\u6574\u7406\u6700\u5168\u7684\u4e00\u4e2abenchmark<\/strong>.<\/p>\n\n\n\n

      https:\/\/microsoft.github.io\/CodeXGLUE\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

      Codex https:\/\/openai.com\/blog\/openai-codex\/ Evaluating … \u7ee7\u7eed\u9605\u8bfbOpenAI \u4ee3\u7801\u751f\u6210\u6a21\u578b Codex\uff1a Evaluating Large Language Models Trained on Code<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[21,4,9],"tags":[],"_links":{"self":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/11428"}],"collection":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/comments?post=11428"}],"version-history":[{"count":60,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/11428\/revisions"}],"predecessor-version":[{"id":11537,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/posts\/11428\/revisions\/11537"}],"wp:attachment":[{"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/media?parent=11428"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/categories?post=11428"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/139.9.1.231\/index.php\/wp-json\/wp\/v2\/tags?post=11428"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}