Isku dhafka Kediska Khubarada

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Isku-dhafka khubarada, maxaa go'aaminaya khabiirada farsameeya calaamada la bixiyay?
   • Router yar oo la bartay (shabakadda gating) oo buundooyinka khubarada
   • Miis-fiirin go'an oo la dejiyay tababarka ka hor
   • Dhererka degdegga isticmaalaha
   • Bakhtiyaa nasiib ah tallaabo kasta
2. Muxuu moodelka MoE u yeelan karaa cabbirro aad uga badan wadarta guud marka loo eego qaabka cufan iyada oo aan la kordhin qiimaha qiimaha calaamad kasta?
   • Waxay ku cadaadisaa halbeegyada hal miisaan oo la wadaago
   • Waxay gabi ahaanba ka booddaa lakabyada dareenka
   • Kaliya dhawr khabiir (top-k) ayaa loo hawlgeliyay calaamad kasta
   • Waxay maamushaa khabiir kasta laakiin celceliska natiijooyinka
3. Dhibaato noocee ah ayay wax ka qabataa isu dheelitirka culeyska (kaaliyaha) khasaaraha inta lagu jiro tababarka MoE?
   • Calaamadaha oo aad ugu dheer daaqadda macnaha guud
   • Router-ka ayaa si xad dhaaf ah u isticmaalaya khubaro yar oo gaajoonaysa inta kale
   • Gradients oo ku lumaya madaxa dareenka
   • Ku-habboonaanta habka ansaxinta
4. Mixtral 8x7B waxa ay hawlgelisaa 2 ka mid ah 8deeda khabiir calaamadiiba. Maxay tani ka dhigan tahay xisaabinteeda iyo cabbirkeeda?
   • Waxay isticmaashaa dhammaan xisaabinta khubarada 8 ee calaamad kasta
   • Waxa ay leedahay cabbirro guud oo ka yar 7B qaab cufan
   • gabi ahaanba kuma shaqayn karo GPU-ga
   • Xisaabiyeheeda firfircoon ee calaamad kasta aad ayuu uga yar yahay marka loo eego wadarta cabbirkeeda
5. Why can a sparse MoE still have substantial memory or storage requirements?
   • Every token necessarily activates all experts
   • Inactive experts cease to have parameters
   • Inactive expert weights still need accessible storage even when not selected for a token
   • The router removes all attention layers
6. Two selected MoE expert outputs are 2 and 6, with combination weights 0.25 and 0.75. What is their weighted output?
   • 5
   • 8
   • 2
   • 0.5
7. Increasing an MoE’s expert count keeps the same top-k and expert size. What can still increase serving cost?
   • The selected expert count must automatically become the total count
   • The model must stop using attention
   • Every inactive weight is automatically deleted
   • Router or communication overhead and the storage required for more experts
8. A sparse MoE calls a learned feed-forward subnetwork an expert. What does that name establish?
   • It is certified to answer only one academic discipline
   • It is one of the networks selected by routing, without proving a named subject specialty
   • It must be a human-written collection of rules
   • It contains no learned parameters