{"id":60006,"date":"2021-04-08T13:05:02","date_gmt":"2021-04-08T04:05:02","guid":{"rendered":"https:\/\/smilegate.ai\/?p=60006"},"modified":"2021-04-08T13:18:49","modified_gmt":"2021-04-08T04:18:49","slug":"gpt-neo","status":"publish","type":"post","link":"https:\/\/smilegate.ai\/cn\/2021\/04\/08\/gpt-neo\/","title":{"rendered":"GPT-\u65b0\u5f00\u6e90GPT-3\u9879\u76ee"},"content":{"rendered":"
OpenAI\u7684GPT-3\u662f\u4e00\u79cd\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff0c\u53c2\u6570\u8ba1\u6570\u9ad8\u8fbe175B\u3002\u5c3d\u7ba1GPT-3\u663e\u793a\u4e86\u60ca\u4eba\u7684\u7ed3\u679c\uff0c\u4f46\u5b83\u4e0d\u662f\u5f00\u6e90\u7684\uff0c\u56e0\u6b64\uff0c\u5982\u679c\u60a8\u8981\u5c1d\u8bd5\uff0c\u53ef\u4ee5\u4f7f\u7528AI Dungeon\uff08https:\/\/play.aidungeon.io\/main\/landing<\/a>\u6216\u54f2\u5b66\u5bb6AI\uff08https:\/\/philosopherai.com\/<\/a>\uff09\u3002\u6b64\u5916\uff0c\u7531\u4e8e\u4e0eMicrosoft\u8fbe\u6210\u72ec\u5bb6\u8bb8\u53ef\u534f\u8bae\uff0c\u5f88\u6709\u53ef\u80fd\u5728\u5c06\u6765\u4ed8\u6b3e\u3002<\/p>\n\n\n\n \u975e\u8425\u5229\u6027\u5f00\u6e90\u7814\u7a76\u7ec4\u7ec7Eleuther AI\u53d1\u5e03\u7684GPT-Neo\u662f\u4f7f\u7528GPT-3\u7684\u7ed3\u6784\u5b66\u4e60\u7684\u4e00\u79cd\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff0c\u8fd8\u53d1\u5e03\u4e86\u6570\u636e\u96c6\u548c\u9884\u5148\u8bad\u7ec3\u7684\u6a21\u578b\u3002\u4ee5\u4e0b\u662f\u6307\u5411GPT-Neo\u548cPipele\u7684github\u5b58\u50a8\u5e93\u7684\u94fe\u63a5\uff1a<\/p>\n\n\n\n GPT-Neo\u57fa\u4e8e\u7f51\u683c\u5f20\u91cf\u6d41\uff0c\u5927\u578b\u5e76\u884c\u5b66\u4e60\u5e93\u521b\u5efa\uff0c\u5e76\u516c\u5f00\u4e86\u5177\u67091.3B\u53c2\u6570\u7684\u6a21\u578b\u548c\u5177\u67092.7B\u53c2\u6570\u7684\u6a21\u578b\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u3002\u6b64\u5916\uff0cGPT-Neo\u5df2\u6dfb\u52a0\u5230HuggingFace\u4e2d\uff0c\u4f7f\u5176\u6613\u4e8e\u4f7f\u7528\u3002\u4ee5\u4e0b\u662fHuggingFace\u7684GPT-Neo\u94fe\u63a5\uff0c\u5e76\u63d0\u4f9b\u4e86\u5176\u4ed6\u5177\u6709125M\u548c350M\u53c2\u6570\u7684\u578b\u53f7\uff1a\u56db\u79cd\u578b\u53f7\u7684GPT-Neo 125M\uff0cGPT-Neo 350M\uff0cGPT-Neo 1.3B\u548cGPT-Neo 2.7B\uff0c\u60a8\u53ef\u4ee5\u5c1d\u8bd5\u4e00\u4e0b\u3002<\/p>\n\n\n\n \u540c\u65f6\uff0cEleuther AI\u4e5f\u6b63\u5728\u5f00\u53d1GPT-Neo\u7684\u540e\u7eed\u9879\u76eeGPT-NeoX\u3002\u4e0e\u57fa\u4e8e\u7f51\u683c\u5f20\u91cf\u6d41\u7684GPT-Neo\u4e0d\u540c\uff0cGPT-NeoX\u5177\u6709NVidia\u5a01\u9707\u5929\u548cDeepSpeed\uff08https:\/\/smilegate.ai\/2021\/01\/27\/deepspeed-fairscale\/<\/a>\uff09\u5e76\u5c06\u4ee3\u7801\u5e93\u79fb\u81f3pytorch\u800c\u4e0d\u662ftensorflow\u3002\u6839\u636eEleuther AI\u7684\u8bf4\u6cd5\uff0c\u6211\u4eec\u8ba1\u5212\u6700\u7ec8\u8bad\u7ec3\u51fa\u4e00\u4e2a\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u5177\u6709\u8bb8\u591a\u4e0eGPT-3 175B\u76f8\u4f3c\u7684\u53c2\u6570\uff0c\u56e0\u6b64\uff0c\u6709\u671b\u8fdb\u884c\u5404\u79cd\u5176\u4ed6\u5206\u6790\u548c\u5e94\u7528\u3002\u8fd9\u662fGPT-NeoX\u7684github\u5b58\u50a8\u5e93\u7684\u94fe\u63a5\u3002<\/p>\n\n\n\n<\/div><\/div>
<\/div><\/div>