Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo cyo Guhindura Ikibazo

Shakisha ibimenyetso bya DiT, gutondekanya, gupima, videwo umwanya-wo kwerekana, hamwe nigiciro cyo kwitondera ibimenyetso byinshi.

Inzira zijyanye no kuyoboraImpinduka
Ikibazo 1 Bya 8

Niki kintu gihindura Diffusion Transformer ugereranije na moderi gakondo yo gukwirakwiza?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Niki kintu gihindura Diffusion Transformer ugereranije na moderi gakondo yo gukwirakwiza?
    • Icyemezo cya U-Net
    • Kode ya kodegisi
    • Ishusho ya decoder
    • Gahunda y'urusaku
  2. Nigute DiT ihindura ishusho mubintu Transformer ishobora gutunganya?
    • Ihindura pigiseli mu magambo
    • Igabanya ishusho yihishe mubice bifatwa nkibimenyetso
    • Ikoresha ikimenyetso kimwe kumashusho yose
    • Ihindura ishusho kumajwi mbere
  3. Ni iki impapuro z'umwimerere DiT zabonye zijyanye no gupima?
    • Imikorere ya plateaus vuba nubunini
    • Moderi nini burigihe ikora nabi
    • Imikorere itezimbere nkuko ubunini bwikitegererezo bugenda bwiyongera kandi ingano ya patch igabanuka
    • Ingano yububiko nta ngaruka igira
  4. Nigute DiTs itera inshinge nka timestep hamwe nibisobanuro byihuse?
    • Mugushyiramo nka pigiseli yinyongera
    • Mugutoza icyitegererezo buri ntambwe
    • Binyuze mu ishami ryihariye
    • Binyuze muburyo busanzwe bwo guhuza n'imiterere (adaLN)
  5. Ni iki gutangiza 'adaLN-zeru' bigeraho?
    • Bituma buri gihinduramatwara itangira nkigikorwa cyiranga imyitozo ihamye
    • Isiba ibice bisanzwe
    • Ihagarika icyitegererezo cyose
    • Ihinduranya ishusho yinjiza
  6. Nubuhe buryo buzwi bukoresha umugongo wa Transformer mugihe cyigihe cyo gukora amashusho?
    • GAN gakondo
    • __AIU_PROTECTED_10 __ 'Sora
    • Umuyoboro usubiramo
    • Icyitegererezo cyibiti
  7. Why can full self-attention become costly when a DiT processes many patches?
    • Its cost stays constant for any token count
    • It must retrain the U-Net at each denoising step
    • Its pairwise token comparisons grow quadratically with the number of tokens
    • It requires a separate language model for every patch
  8. After a DiT processes latent image patches with its Transformer blocks, what happens to the patch outputs?
    • As one text token for an entire video
    • As only a sequence of audio waveforms
    • As a separate convolutional U-Net for each frame
    • They are reassembled and decoded back into pixels