{"id":59811,"date":"2021-01-27T10:22:51","date_gmt":"2021-01-27T01:22:51","guid":{"rendered":"https:\/\/smilegate.ai\/?p=59811"},"modified":"2021-01-27T10:22:53","modified_gmt":"2021-01-27T01:22:53","slug":"deepspeed-fairscale","status":"publish","type":"post","link":"https:\/\/smilegate.ai\/cn\/2021\/01\/27\/deepspeed-fairscale\/","title":{"rendered":"\u5927\u578b\u6a21\u578b\u7684\u5185\u5b58\u4f18\u5316\u6280\u672f"},"content":{"rendered":"

\u968f\u7740\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u7684\u53c2\u6570\u6570\u91cf\u663e\u7740\u589e\u52a0\uff0c\u8bad\u7ec3\u6240\u9700\u7684\u5185\u5b58\u4e5f\u968f\u4e4b\u589e\u52a0\u3002 OpenAI\u7684GPT-2\u75311.5B\u53c2\u6570\u7ec4\u6210\uff0c\u800cGoogle\u7684mT5\u4e5f\u5177\u670913B\u53c2\u6570\u3002\u53e6\u5916\uff0cOpenAI\u7684GPT-3\u7684\u53c2\u6570\u6570\u91cf\u8fbe\u5230175B\u3002\u5728\u8fd9\u79cd\u5927\u578b\u53f7\u7684\u60c5\u51b5\u4e0b\uff0c\u5373\u4f7f\u5c1d\u8bd5\u4f7f\u7528\u5c0f\u6570\u636e\u8fdb\u884c\u5fae\u8c03\uff0c\u4e5f\u5f88\u96be\u8f7b\u677e\u5730\u5c1d\u8bd5\u5b83\uff0c\u56e0\u4e3a\u5b83\u9700\u8981\u5927\u91cf\u7684GPU\u5185\u5b58\u3002<\/p>\n\n\n\n

\u5fae\u8f6f\u53d1\u8868\u7684\u4e00\u7bc7\u8bba\u6587\u300a ZeRO\uff1a\u8bad\u7ec3\u4e07\u4ebf\u53c2\u6570\u6a21\u578b\u7684\u5185\u5b58\u4f18\u5316\u300b\u6db5\u76d6\u4e86\u7f13\u89e3\u6b64\u95ee\u9898\u7684\u6280\u672f\uff0c\u5e76\u4e14\u5df2\u77e5\u53ef\u4ee5\u663e\u7740\u51cf\u5c11\u6240\u9700\u7684\u6700\u5927\u5185\u5b58\u91cf\u3002\u8fd9\u662f\u8bba\u6587\u7684\u94fe\u63a5\uff1a<\/p>\n\n\n\n

<\/div>\n\n\n