Kedis-ka-soocidda is-khilaafsan

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Decoding ka duwan, doorkee ayuu ka ciyaaraa moodelka 'amateur'?
   • Itimaa-galnimadeeda log-ka waa la kala jaray si loo xakameeyo hababka fashilka la wadaago
   • Waxay si toos ah u soo saartaa qoraalka ugu dambeeya
   • Waxaa si fiican loo habeeyey khaladaadka khabiirka
   • Waxay beddeshaa khabiirka marka khabiirku aanu hubin
2. Maxay tahay sababta kala-goynta hiwaayadda ixtimaalka u yaraynaysaa soo-noqoshada iyo wax-soo-saarka xun?
   • Hiwaayadda waxaa loo tababbaray inuusan waligiis ku celin erayada
   • Calaamadaha guud iyo kuwa soo noqnoqda ayaa doorbidaya labada nooc, markaa ka duwanaanshuhu wuu baabi'iyaa
   • Hiwaayaddahu waxa ay tirtirtaa calaamadaha nuqul ka mid ah jiilka ka dambeeya
   • Ku celcelintu waa wax aan suurtagal ahayn mar haddii laba nooc ay codeeyaan
3. Maxay qabataa xayiraadda la qabsiga macquulnimada ee 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. Qiyaastii waa maxay qaacidada dhibcaha ee dejinta kala duwan?
   • log p_amateur laga jaray log p_expert
   • waqtiyada khabiirka p_amateur
   • Celceliska labada itimaalka
   • Log p_expert laga jaray waqti isku-xirnaan ah log p_amateur
5. Faa'iidada muhiimka ah ee wax ku oolka ah ee dejinta isbarbardhigga waa in ay:
   • Wuxuu shaqeeyaa waqti go'an oo aan lahayn tababar dheeraad ah
   • Waxay u baahan tahay dib u tababarida labada nooc si wadajir ah
   • Kaliya waxa uu ka shaqeeyaa hawlaha kala soocida
   • Waxay meesha ka saaraysaa baahida loo qabo nooc weyn
6. Qaabkee dambe u fidiya fikradda isbarbardhigga ilaa lakabyada hal nooc?
   • Baadhitaanka Beam
   • DoLa, kaas oo ka soo horjeeda hore iyo lakabyada dambe
   • Codaynta hunguriga
   • Heerkulka oo kor u kaca
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