Isbarbar-dhigga Khabiirka ee MoE U Adeegista Kediska

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Waa maxay isbarbardhigga khabiiradu u qaybiyaan GPU-yada oo dhan?
   • Nuqullada moodelka buuxa ee shaqo-joojinta
   • Feejignaanta qofka ayaa kaliya
   • Khubaro kala duwan oo soo jeedinta quudinta ee lakabka MoE
   • Lakabyada miiska gundhigga
2. Qaabkee isgaarsiineed ayaa udub dhexaad u ah isbarbardhigga khabiirada MoE?
   • Laba wax-isweydaarsiga oo dhan (dir iyo isku dar)
   • Hal baahin
   • Isku-dubbarid jaan-ilaa- dhibic
   • Looma baahna isgaarsiin
3. Waa maxay sababta MoE ay u xisaabiso hoos-u-dhac kasta iyadoo ay jirto wadarta guud?
   • Waxay ku cadaadisaa miisaanka ilaa 1 bit
   • Waxay ka booddaa lakabyada dareenka
   • Khubaradu waxay wadaagaan dhammaan miisaankooda
   • Kaliya khabiirada ugu sarreeya ee la doortay ayaa ku shaqeeya halkii calaamad
4. Dhibaato noocee ah ayaa soo baxda sababtoo ah marin-ku-gudbinta waxay ku tiirsan tahay xog?
   • Qarax yar
   • Isku dheelitir la'aanta culeyska, halkaasoo khubarada qaar ay ka helaan calaamado aad u badan
   • Router-ka lama tababari karo
   • Calaamaduhu waxay lumiyaan booskooda
5. Why are capacity limits and padding used in some MoE implementations?
   • To control expert buffer sizes and make computation shapes more predictable
   • To require every expert to process every token
   • To eliminate all token routing
   • To guarantee that every MoE system drops tokens
6. Sidee bay nidaamyadu u qariyaan daahitaanka talaabooyinka-dhan-dhan?
   • Adigoo ku shaqeeya CPU beddelkeeda
   • Adiga oo meesha ka saaraya talaabada isku dhafka ah
   • Iyadoo la kordhinayo heerka waxbarashada
   • Iyadoo la isku dhejinayo isgaarsiin la leh xisaabinta khabiirada
7. A routed token chooses two MoE experts on different devices. What must expert-parallel execution do?
   • Run every expert in the model regardless of routing
   • Dispatch its representation to the selected experts and combine their outputs
   • Replace both expert networks with an attention head
   • Drop the token solely because it selected two devices
8. One MoE expert receives far more tokens than others. Why can the layer’s latency increase?
   • The total expert parameter count automatically doubles
   • The router’s outputs become attention positions
   • The overloaded expert becomes a straggler while other devices wait
   • All other experts are removed from memory