Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo cyo Gutandukanya Ibinyuranyo

Ikizamini cyinzobere - amanota yikinamico, gusangira-token guhagarika, gushungura byoroshye, gushungura-gukoresha igihe, no gusuzuma ibisekuruza bitandukanye.

Inzira zijyanye no kuyoboraGutandukanya ibintu
Ikibazo 1 Bya 8

Mu buryo butandukanye bwo gushushanya, ni uruhe ruhare ruto 'rwikinisha' rufite?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Mu buryo butandukanye bwo gushushanya, ni uruhe ruhare ruto 'rwikinisha' rufite?
    • Log-probabilite yakuweho kugirango bahagarike uburyo bwo gutsindwa
    • Itanga inyandiko yanyuma isohoka muburyo butaziguye
    • Byakosowe neza kumakosa yinzobere
    • Isimbuza umuhanga mugihe impuguke idashidikanywaho
  2. Ni ukubera iki gukuramo amahirwe yo kwikinisha bigabanya gusubiramo no gusohora bland?
    • Amateur yatojwe kutazongera gusubiramo amagambo
    • Ibimenyetso rusange kandi bisubirwamo bitoneshwa na moderi zombi, bityo itandukaniro rirabasesa
    • Amateur asiba ibimenyetso byigana nyuma yibisekuru
    • Gusubiramo ntibishoboka iyo moderi ebyiri zitoye
  3. What does the adaptive plausibility constraint do in contrastive decoding?
    • It guarantees every remaining continuation is coherent and true
    • It replaces the expert distribution with the amateur distribution
    • It filters out candidates whose expert probability is too low relative to the expert’s top token
    • It forces every token to have equal probability
  4. Hafi ya formula yo gutanga amanota muri decoding itandukanye?
    • log p_amateur ukuyemo log p_umuhanga
    • p_ibihe byiza p_amateur
    • Impuzandengo y'ibishoboka byombi
    • injira p_ubushakashatsi ukuyemo inshuro coefficient log p_amateur
  5. Inyungu yingenzi ifatika yo gutandukanya decode ni uko:
    • Akora mugihe cyumwanzuro nta mahugurwa yinyongera
    • Irasaba kongera imyitozo yombi hamwe
    • Gusa ikora kumirimo yo gutondeka
    • Kurandura gukenera icyitegererezo kinini
  6. Nubuhe buryo bwanyuma bwagura igitekerezo gitandukanye kurwego rumwe rwihariye?
    • Gushakisha ibiti
    • DoLa, itandukanye hakiri kare nubukererwe
    • Umururumba
    • Igipimo cy'ubushyuhe
  7. The expert assigns probability 0.4 to its top token and the plausibility threshold is 0.1 of that value. What is the minimum probability for a candidate to remain eligible?
    • 0.4
    • 0.1
    • 0.04
    • 0.004
  8. Why can contrastive decoding require more inference work than using the expert model alone?
    • It must retrain both models for every prompt
    • It replaces the vocabulary with image patches
    • It runs only the amateur model
    • It scores candidates with both an expert and an amateur model at inference time