Kediska Ganacsiga Dib u Xisaabinta Hawlgelinta

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Muxuu dib u xisaabinta hawlgelintu ka beddelaa kaydinta xusuusta?
   • Saxnaanta moodeelka
   • Xisaabinta dheeraadka ah (FLOPs)
   • bandwidth network
   • Saxnaanta tirada
2. Waa maxay sababta dhaqdhaqaaqa gudbinta-gudbigu si caadi ah u kaydsan yihiin?
   • Si loo tuso isticmaalaha
   • Si loo cadaadiyo qaabka
   • Waxay u baahan yihiin calaamadaynta
   • Sababtoo ah faafinta dhabarku waxay u baahan yihiin inay xisaabiyaan gradients
3. Why should activation checkpointing overhead be measured on the actual model?
   • Checkpoint placement, operators, and hardware determine the cost of replay
   • It always adds exactly forty percent to wall-clock time
   • It guarantees a tenfold speed improvement
   • It removes all forward computation
4. For an idealized N-layer chain with comparable activations, what storage scaling can square-root checkpoint placement achieve?
   • O(N squared) for all model parameters
   • O(1) for all training memory regardless of N
   • O(sqrt(N)) for activation storage
   • O(N squared) for optimizer state
5. Waa maxay fikradda ka dambeysa dib u xisaabinta xulashada (aan buuxa ahayn)?
   • Dib u xisaabi qalliin kasta laba jeer
   • Weligaa waxba ha xisaabin
   • Dib u xisaabi oo kaliya lakabka gundhigga
   • Dib u xisaabi kaliya xusuusta-culus laakiin xisaabin-jaban ops, kayd qaali ah
6. Waa kuwee farsamada dhameystirka ah ee inta badan lagu daro xisaabinta si loo badbaadiyo xitaa xasuus badan?
   • Kordhinta cabbirka dufcadda
   • U dajinta hawlgelinta CPU ama NVMe
   • Ka saarida dhammaan lakabyada caadiga ah
   • U beddelashada habka wax-ka-sheegga
7. A checkpointed block contains dropout. Why does consistent RNG state matter during replay?
   • Dropout becomes a model-weight quantizer
   • The optimizer no longer needs gradients
   • Backward recomputation should reproduce the intended forward randomness
   • The batch size must become one
8. Checkpointing reduces stored activations, but optimizer states still exceed GPU memory. Why is the job still out of memory?
   • Activation recomputation does not remove optimizer-state storage
   • Checkpointing increases every optimizer tensor to FP64
   • Backward no longer consumes any memory
   • The model cannot use checkpointing with an optimizer