Kedis ku xidhan hagaha · Adag Heerka

Muunad la qorsheeyay iyo Kediska Eexda Soo Bandhigida

Waxay hubisaa sida muunad jadwalaysan u beddesho macnaha guud, sababta ixtimaalka summada dahabka ahi u qudhunto, iyo isbeddellada is-dhaafsiga ee muunada.

Wadooyinka haga ee la xidhiidhaMuunad la qorsheeyay iyo Eexda Soo Bandhigista
Su'aal 1 ee 8

Eexda soo bandhigista ayaa soo baxda sababtoo ah moodooyinka loo tababaray macalinka oo ku qasbaya shardi kaliya:

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Eexda soo bandhigista ayaa soo baxda sababtoo ah moodooyinka loo tababaray macalinka oo ku qasbaya shardi kaliya:
    • Calaamadaha dhabta ah ee dhabta ah, laakiin waa inay adeegsadaan tooda marka la eego
    • Calaamadahooda la saadaaliyay
    • Calaamadaha qaylada random
    • Calaamadaha mustaqbalka
  2. Muxuu sameeyaa muunad jadwalaysan inta lagu jiro tababarka?
    • Had iyo jeer quudiya calaamadda runta ah
    • Si aan kala sooc lahayn u kala doorta calaamadda runta ah iyo saadaasha moodeelka
    • Wuxuu curyaamiyaa shaqada luminta si xilliyo ah
    • Waxay labanlaabtaa heerka waxbarashada waqti ka dib
  3. Sidee bay suurtogalnimada isticmaalka calaamadda runta dhabta ah isu beddeshaa tababarka muunad la qorsheeyay?
    • Waa joogto
    • Hoos ayay ku bilaabataa wayna korodhaa
    • Waxay ku bilaabataa sare waxayna u qudhuntaa ilaa eber
    • Si aan kala sooc lahayn ayay u ruxaysaa
  4. Yaa soo bandhigay muunad la qorsheeyay, iyo qiyaas ahaan goorma?
    • Vaswani iyo al., 2017
    • Hinton iyo al., 2006
    • LeCun iyo al., 1998
    • Bengio et al., 2015
  5. Dib u dhac farsamo oo la aqoonsan yahay ee quudinta calaamadihii muunadyada ahaa ee dib loogu soo celiyay muddada tababarka ayaa ah:
    • Talaabada muunada gaarka ah waa mid aan la kala sooci karin, oo adkaynaysa gradients
    • Waxay ka dhigtaa tababarka mid si buuxda loo go'aamiyo
    • Uma baahna xog xaqiijin
    • Waxay meesha ka saartaa baahida loo qabo in la isku dhejiyo
  6. Jadwalka qaabkee ayaa inta badan loo soo qaatay inuu burburiyo suurtagalnimada qasabka macalinka?
    • Joogto ah
    • Sigmoid suus ah
    • Kor u kaca toosan ee kor u kaca
    • Socod random
  7. At a decoding step where scheduled sampling chooses the model-output branch, what is fed as the next context token?
    • The next ground-truth token every time
    • A token from a separate validation sequence
    • A token sampled from the model’s predicted distribution
    • The average of all vocabulary embeddings
  8. Why does scheduled sampling usually begin with a high probability of using ground truth?
    • It guarantees unbiased gradients
    • It removes the need for a training schedule
    • It makes inference use ground-truth tokens
    • Early predictions are poor, so a correct prefix can stabilize initial learning