quiz bu lëkkale ak tegtal · Dëgër Niveau

Optimiser dafay wàcce CPU ak NVMe Quiz

Jëfandikool optimisatër biy dindi kontabilite mémoire, plasement yeesal host, ak toxal kompromis, boole ci li tax dindi du dindi bépp GPU allocation.

Yooni njiit yu jëm ciOptimiser bi dafay wàcce ci Cpu ak Nvme
Laaj 1 de 8

Lan mooy mébet bu njëkk bi ñuy dindi etaa optimiser yi ci CPU wala NVMe?

Xoolaatal laaj bu nekk8 laaj

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

  1. Lan mooy mébet bu njëkk bi ñuy dindi etaa optimiser yi ci CPU wala NVMe?
    • Ngir defar model bi mu gëna dëggu
    • Wàññi limu jamonoy tàggat yaram yiñ soxla
    • Ngir dindi soxla gradient yi
    • Ngir méngoo ak model yu gëna mag ci dindi mémoire GPU bu néew
  2. For an Adam setup with two FP32 moments and an FP32 master weight, how much memory do those three values use per parameter?
    • Two bytes for all three values
    • Twelve bytes, excluding model weights and gradients
    • Sixteen bytes for the three values alone
    • Zero bytes because they are implicit
  3. Ban kaadar moo siiwal dechargement optimiser bu mag?
    • Fonction aktivasioŋ bu bees
    • Kompresioŋ nataal JPEG
    • DeepSpeed ZER-Decharge ak ZER-Infini
    • Dagg degrade
  4. Lan mooy njëgu performance bi gëna mag ci wàcce ci CPU wala NVMe?
    • Done yi dañu wara jaar ci lëkkalekaay yu yeex yu melni PCIe wala ci disk bi, loolu dafay yokk latency
    • Royuwaay bi fàtte na tàggat yaram bu njëkk
    • Optimiser bi defay juum ci wàllu math
    • GPU tàngoor wàccina lool
  5. naka la sistemu decharge di nëbbe lu bari ci latency transfert bi?
    • Soo sànnee paas bi ci ginaaw
    • Ni ñuy jaxasoo prefetching done ak xayma GPU biy wéy
    • Soo tàggatee benn parametre ci saa si
    • Ni ñuy dindi njubte bu jaxaso
  6. Ci ZeRO-Offload, fan la Adam di faral di def?
    • lëmm ci kaw njiitu NVMe
    • Ci GPU bu wuute
    • Ci kaw CPU bi, jëfandikoo poids master ak moment yiñ denc foofu
    • Ci biir kommutatëru reso bi
  7. Optimizer states are offloaded, but GPU weights and activations still exceed capacity. What follows?
    • The job can still run out of GPU memory
    • Offloading guarantees the job fits any GPU
    • The optimizer moments automatically become model activations
    • The model no longer needs forward tensors
  8. A CPU-offloaded optimizer becomes the training bottleneck. Which measurement is most relevant?
    • Only the inference accuracy
    • The UI font used for training logs
    • Only the number of dataset classes
    • Host optimizer-update time and host-device transfer time