Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo cya GPTQ na AWQ Nyuma yo Kwibaza Amahugurwa

Kugenzura itandukaniro riri hagati ya GPTQ na AWQ, kalibrasi yamakuru, indishyi zagabanijwe, imiyoboro-imenyekanisha ibikorwa hamwe nububiko buke.

Inzira zijyanye no kuyoboraGptq na Awq Post Post Quantization
Ikibazo 1 Bya 8

Bisobanura iki 'kugereranya nyuma y'amahugurwa' bisobanura?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Bisobanura iki 'kugereranya nyuma y'amahugurwa' bisobanura?
    • Gucomeka uburemere bwicyitegererezo kimaze gutozwa kugirango ugabanye neza neza utarangije imyitozo yuzuye
    • Kongera imyitozo rwose muburyo buke
    • Ongeraho andi makuru yamahugurwa nyuma yo koherezwa
    • Kongera icyitegererezo cyibara ryibara
  2. Ni ayahe makuru GPTQ ikoresha kugirango yishyure amakosa yo kuzenguruka mugihe cyo kubara?
    • Urutonde rwamagambo
    • Icyiciro cya kabiri (Hessian / curvature) amakuru avuye muri kalibrasi
    • Umukoresha amateka yihuse
    • Urusaku rusanzwe
  3. Ni ubuhe bushishozi bwingenzi butwara AWQ (Gukora-Kumenya Ibiro)?
    • Ibipimo byose ni ngombwa
    • Ibikorwa bigomba kubarwa ariko uburemere busigaye budakozweho
    • Agace gato k'imiyoboro yuburemere ni ngombwa cyane kandi igomba kurindwa
    • Quantisation igomba kubaho mugihe cyo kwitegura gusa
  4. Hafi yangahe yibuka bingana na 4-bit yo kubika hamwe nuburemere bwa 16-bit?
    • Nta kuzigama
    • Hafi inshuro ebyiri kwibuka
    • Kimwe cya kabiri cyo kwibuka
    • Hafi ya kimwe cya kane cyo kwibuka (kugabanya 4x)
  5. Ni ukubera iki GPTQ na AWQ byombi bigereranya uburemere ariko bikagumya gukora neza?
    • Ibiro byiganje mu kwibuka, kandi kugereranya ibikorwa bikabije bikomeretsa ukuri kurushaho
    • Ibikorwa ntibibaho muri transformateur
    • Ibikorwa bimaze kuba 1-bit
    • Birasabwa n'amategeko
  6. Niki gikoresho gikunze gukoreshwa mugukoresha moderi zingana mubikoresho byabaguzi?
    • Microsoft Ijambo
    • llama.cpp
    • Urupapuro rwerekana urupapuro
    • Urubuga rwonyine
  7. How does GPTQ use a calibration dataset during layer-wise weight quantization?
    • It adds new training labels to the model
    • It selects a new tokenizer vocabulary
    • It estimates second-order information to compensate for rounding errors
    • It measures only GPU temperature
  8. How does AWQ identify which weight channels deserve protection?
    • It computes a full Hessian inverse
    • It adds random noise to all weights
    • It selects channels only by their index
    • It uses activation magnitudes to identify channels important to preserve