ã¯ããã« Turing æ ªå¼ä¼ç¤¾ã®ãªãµã¼ããã¼ã ã§ã¤ã³ã¿ã¼ã³ããã¦ããæ±äº¬å·¥æ¥å¤§å¦ B4 横ç°ç 究室ã®è¤äº(@okoge_kaz)ã§ãã èªç¶è¨èªå¦çåéã«ããã大è¦æ¨¡æ·±å±¤å¦ç¿ã®éè¦æ§ã¯æ¥ã«æ¥ã«é«ã¾ã£ã¦ãã¾ãããGPT-3, GPT-4 ãªã©ã®ã¢ãã«ã®å¦ç¿ã«ã¯è¨å¤§ãªè¨ç®ã³ã¹ãããããã容æã«å¦ç¿ã§ããªããªã£ã¦ãã¾ããå®éãã¢ãã«ãµã¤ãºãè¿å¹´æ¥éã«å¤§ãããªã£ã¦ãããã¨ã«ãããå¦ç¿ã«å¿ è¦ãªè¨ç®é(FLOPs)ã¯ä»¥ä¸ã®ããã«å¹´ã è¨å¤§ã«ãªã£ã¦ãã¾ããè¿å¹´ã®å¤§è¦æ¨¡ã¢ãã«ã§ã¯ãNVIDIA H100 80GB ã§ãã£ã¦ã 1 ã¤ã® GPU ã§ã¯ãã¢ãã«ãã®ãããã¨ããã§ãã¾ããã Compute Trends Across Three Eras of Machine Learning ãã ã¾ãScaling Laws ã«ããã¨ã大è¦æ¨¡ãªã¢ãã«ã¯å°ããã¢ãã«ã¨æ¯è¼ãã¦ããåªããæ§è½ãçºæ®ãããããèªå
{{#tags}}- {{label}}
{{/tags}}