ååä¸ã¯è«æèªã®æ»èªã³ã¡ã³ãã¸ã®å¯¾å¿çãåè°ãããã¼ãã£ã³ã°ãæè¿ã¯ãã¾ãè«æèªã«æç¨¿ãã¦ããªãï¼å¹´ã«1-2æ¬ãããï¼ã®ã§ãããããæéãããã¦ã³ã¡ã³ããããã£ã¦ãããã¦ããç³»ã®ç ç©¶ã¹ã¿ã¤ã«ã¯åãã¦ããªãã®ã ãããã¡ã®ãããªæå¡1人ã®ç 究室ã ã¨ãçè·é¢èµ°ã®å½éä¼è°ãããé·è·é¢èµ°ã®è«æèªã®æ¹ãç ç©¶ãµã¤ã¯ã«ã¨ãã¦åãã¦ããã®ããªã¨æã£ãããããM2 ã®å¦çã¯ç¹ã« M2 ã®ç§ãããã¾ã§ã«æç¨¿ãã¦ããã¦ãä¿®å£«è«æã®ä»ä¸ãã¨è¢«ãã¦ããã¨ã¡ããã©ããã®ããªã¨æã£ããã
åå¾ã¯è«æç´¹ä»ãèããã¹ã©ã¤ã
- Yun Chen, Victor O.K. Li, Kyunghyun Cho, Samuel R. Bowman. A Stable and Effective Learning Strategy for Trainable Greedy Decoding. EMNLP 2018.
ããã¯æ§æè§£æã®ç ç©¶ãªã®ã ããæ¬ä¼¼ãã¼ã¿ã使ã£ã¦è¨ç·´ããã¨ãã«ãã©ããããµãã«å ¥ããã¨ããããã¨ãããããªè¨å®ãè²ã ã¨è©¦ãã¦ãããæ¬ä¼¼ãã¼ã¿ã使ã話ã¯è²ã æè¿è©¦ãã¦ããã®ã§ãããããã£ããæ§æè§£æã®è«æã¯ã¾ã 1æ¬ãæ¸ãããã¨ããªãã®ã§ãä¸åº¦æ¸ãã¦ã¿ããã¨ã¯æã£ã¦ããã
夿¹ã¯ç ç©¶ä¼ã«ã¦ä¿®å£«è«æã®é²æå ±åãèããã¨ã¯ãããä»ã®æ©ä¼ã«èãã¦ãã人ãå¤ãã®ã§ãååã®äººã®åã§ããããç 究室ã6å¹´ç®ãä¿®å£«è«æãæ¸ãå¦çã4代ç®ãªã®ã§ã大å質ãéãè½ã¡çãã¦ãã¦ãåéçã«ã¯ã¿ããªã ããã30-40ãã¼ã¸ããããå 容çã«ã¯ 2nd tier ã®å½éä¼è°ã®ãã«ãã¼ãã¼ç¸å½ï¼å°ãããã°ãã°å½å ã®åæè«æèªç¸å½ï¼ãããã®ãã®ãæ¸ããããã«ãªã£ã¦ããã®ã§ããããµã¤ã¯ã«ã«ãªã£ã¦ããã¨æã£ã¦ããããã¡ã®ç 究室ã ãã§ãªãã審æ»å§å¡ã«ãªã£ã¦ããä»ç 究室ã®ä¿®å£«è«æãããªãã¯ãªãªãã£ãä¸ãã£ã¦ããã®ã§ãå°éã«ã³ã¡ã³ãããã¦ããã¨ãå 輩ã®ä¿®å£«è«æãè¦ã¦ã¿ããªçä¼¼ãã¦ãããã®ããªã¨æã£ããããã
ç ç©¶ä¼ã®ãã¨ã«ãµã¼ãã®æ§åãè¦ã«ãããã©ããæ¨æ¥ã®å¤æ¹ãããããæ°ãããµã¼ãã«ãã°ã¤ã³ã§ããªããªã£ã¦ããããã§ãã¿ã¤ã ã¢ã¦ãã®ä»æ¹ããèªè¨¼ï¼LDAPï¼ã§ã¯ãªãããªãã¨æã£ã¦ãããããã¯ã LDAP ãµã¼ãããªãããã®åå ã§æ¢ã¾ã£ã¦ããããã§ãLDAP ãæä¾ãã¦ãã QNAP ãåèµ·åãããç´ã£ããGPU ãµã¼ãã®ããã»ã¹ããã£ã¨åãã¦ãããããªã®ã§ãå¤ã«æ¢ããªãã¦ããã£ããããã QNAP ã¯ç®¡çãç°¡åã ããHDD 以å¤ã«é害ããã£ãã¨ãã«è§£æ±ºããã®ãé£ããæ°ãããï¼ããã¯æ¬æ¥ QNAP ã®ä»äºã§ã¯ãªãã®ã§ã¯ãããï¼ã