quiz bu lëkkale ak tegtal · Diggante Niveau

Laajte ci Transformatëri Diffusion

Saytu jeton patch DiT, kondisioneer, escalier, wideo espace-temps, ak njëgu bàyyi xel ci jeton yu bari.

Yooni njiit yu jëm ciTransformatëru diffusion
Laaj 1 de 8

Lan mooy wecci transformatëru diffusion buñu ko méngale ak xeetu diffusion yiñ gëna xam?

Xoolaatal laaj bu nekk8 laaj

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

  1. Lan mooy wecci transformatëru diffusion buñu ko méngale ak xeetu diffusion yiñ gëna xam?
    • Denoisatëru U-Net bi
    • Encodeur de text
    • Dekodeeru nataal bi
    • Jamonoy bruit bi
  2. naka lay def ba DiT soppi nataal mu nekk lu Transformer mëna def?
    • Dafay soppi pixel yi ci baat
    • Dafay xaaj nataal bu nëbbu bi ay pàcc yuñ jappee ni ay token
    • Dafay jëfandikoo benn token ngir nataal bi yépp
    • Dafay njëkka soppi nataal bi ci audio
  3. Lan la këyitu DiT bi njëkk gis ci wàllu eskalaasioŋ?
    • Plateau performance gaaw ak dayo
    • Modèle yu gëna mag ñooy gëna bon
    • Performance dafay gëna yokk bu model bi di gëna màgg, patch bi di wàññeeku
    • Dayo patch bi amul benn njeexital
  4. naka lay def ba DiTs di dugal ay kondisioneer yu melni timestep ak bind?
    • Soo koy yokk ci yeneen pixel
    • Sooy tàggataat model bi jéego bu nekk
    • Jaaraleko ci bànxaas bu wuute
    • Jaaraleko ci normalisasioŋ buy méngoo (adaLN)
  5. Lan la 'adaLN-zero' di def?
    • Dafay tax bépp bloku Transformer tàmbali nekk fonction dàntite ngir tàggat bu dëgër
    • Dafay efaase ay couche normalisasioŋ
    • Dafay gelé model bi yépp
    • Dafay dindi nataalu dugal bi
  6. Ban sistem bu siiw mooy jëfandikoo yaxu ndigg Transformer ci kaw patch espace-temps ngir defar wideo?
    • GAN cosaan
    • OpenAI's Sora
    • Reseau neuronal buy baaxoo
    • ab xeetu garabu dogal
  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