Maswali yanayounganishwa na mwongozo · Ngumu Kiwango

Mawasiliano ya Pamoja na Maswali ya NCCL

Jaribu majukumu ya kupunguza, kukusanya yote, kutangaza na kupunguza kutawanya, utambulisho wa NCCL, upakiaji wa mtandao, na mfano mdogo wa kupunguza jumla.

Njia za mwongozo zinazohusianaMawasiliano ya Pamoja na Nccl
Swali 1 ya 8

Operesheni ya kupunguza yote inatimiza nini katika mafunzo yaliyosambazwa?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Operesheni ya kupunguza yote inatimiza nini katika mafunzo yaliyosambazwa?
    • Inachanganya thamani kwenye GPU zote na kuipa kila GPU matokeo yaliyounganishwa
    • Hufuta data isiyohitajika kutoka kwa GPU moja
    • Inapunguza kasi ya saa ya GPU ili kuokoa nishati
    • Inatuma data kutoka kwa GPU zote hadi kwenye diski moja ya hifadhi
  2. Je, kifupi cha NCCL kinamaanisha nini?
    • Safu ya Nguzo ya Neural Compute
    • Mantiki ya Kudhibiti Akiba ya Mtandao
    • Maktaba ya Mawasiliano ya Pamoja ya NVIDIA
    • Maktaba ya Kuhesabu Nambari ya Convolution
  3. Why can chunked ring communication be useful for a large tensor?
    • It guarantees constant latency for any number of devices
    • Chunks can be streamed across participating links instead of waiting for one monolithic transfer
    • It makes topology irrelevant
    • It prevents all network contention
  4. Ni operesheni gani ya pamoja inayokusanya kila kipande cha data cha GPU kuwa nakala kamili inayoshikiliwa na GPU zote?
    • Tangaza
    • Punguza-kutawanya
    • Kizuizi
    • Kukusanya wote
  5. What does a supported network reduction offload such as SHARP do?
    • Guarantee exactly half the traffic for every collective
    • Remove the need for participating ranks
    • Perform some reduction work in compatible network hardware
    • Turn all-gather into a local memory copy
  6. Ni muunganisho gani ambao NCCL kawaida hutumia kwa mawasiliano KATI ya seva (nodi)?
    • InfiniBand au RoCE
    • Kebo ya USB
    • Nafasi ya PCIe pekee
    • Wi-Fi
  7. Three ranks contribute scalar values 2, 3 and 5 to a sum all-reduce. What value does each receive?
    • Only its own original value
    • 3 for the middle rank and nothing elsewhere
    • The average 3.33 unless requested separately
    • 10
  8. Which collective combines contributions but distributes reduced pieces instead of the entire reduced buffer to every rank?
    • Kukusanya wote
    • Punguza-kutawanya
    • Tangaza
    • A standalone model checkpoint