Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Lens-ka Logit iyo Kediska Lens-ga La Daawado

Wuxuu hubiyaa akhrinta lakabka dhexe, qaab dhismeedka qulqulka haraaga, turjubaanka lens-ka ee la hagaajiyay iyo xaddidaadda muraayada logit-ka ceyriinka ah.

Wadooyinka haga ee la xidhiidhaLens-ka Logit iyo Lens-ka La-hagaagsan
Su'aal 1 ee 8

Muxuu sameeyaa muraayada logit?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Muxuu sameeyaa muraayada logit?
    • Wuxuu dabaqaa moodeelka ugu dambeeya ee aan la jaan qaadi karin ee u dhexeeya dawladaha qarsoon si loo akhriyo saadaasha lakabkiiba
    • Wuxuu ku daraa lakabyo cusub transformer
    • Waxay ka saartaa dareenka moodeelka
    • Waxay u tarjumaysaa inta u dhaxaysa luqadaha aadanaha
  2. Waa maxay horumarinta muhiimka ah ee lens-ka la hagaajiyay uu ku darayo muraayada logit cayriin?
    • Waxay isticmaashaa eray bixin ka weyn
    • Waxa ay tababartaa khariidad yar oo lakabka affine ah si ay dawladaha qarsoon ugu turjunto saldhigga ugu dambeeya
    • Waxay meesha ka saaraysaa baahida loo qabo wax kasta oo aan la isku xidhin
    • Waxay kaliya ku shaqeysaa moodooyinka sawirka
  3. Qaab noocee ah ayaa ku jira transformers-ka ka dhigaya muraayadahaas suurtagalka ah?
    • filtarrada convolutional
    • Calaamadeeyaha
    • Qulqulka haraaga dheeriga ah ee aan la xirin hadhow ay mashruucisa si loo gashado
    • Heerka waxbarashada wanaajiya
  4. Yaa keenay lenskii asalka ahaa, iyo qiyaas ahaan goorma?
    • DeepMind sanadkii 2015
    • OpenAI sanadka 2018
    • Google sanadka 2012
    • nostalgebraist sanadka 2020
  5. Halkee buu muraayada logitka inta badan ka tusiyaa jawaabta u dambaysa ee 'cristallizing'?
    • Lakabyada dhexe ilaa dambe
    • Kaliya lakabka ugu horreeya
    • Marna, jawaabtu hal mar bay soo baxaysaa
    • Ka baxsan qaabka gebi ahaanba
  6. Maxaa laga yaabaa in muraayada logit-ka ceeriin aan lagu kalsoonaan karin lakabyada hore?
    • Lakabyada hore ma laha neurons
    • Waddamadooda qarsoon ayaan weli la jaanqaadin wax-soo-saarka ugu dambeeya ee la filayo
    • Lakabyada hore weligood looma isticmaalo saadaasha
    • Matrixka aan xidhnayn waa random
  7. What does the model’s final unembedding map a residual-stream vector into?
    • A new set of transformer layers
    • A similarity graph over training examples
    • Vocabulary logits that can be converted into token probabilities
    • A gradient clipping threshold
  8. Why can an early-layer raw logit-lens prediction be misleading?
    • The layer has already generated the full answer
    • The tokenizer has been removed
    • The unembedding only works on images
    • The intermediate state may not yet be aligned with the final output basis