Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Ciqaabta ku celiska ah iyo Kediska Koontaroolka Go'aaminta

Is barbar dhig heerkulka, top-k, top-p, iyo ganaaxyada soo noqnoqda, oo ay ku jiraan sida logisku isu beddelo ka hor inta aan la samayn muunad socodka shaqada jiilka LLM.

Wadooyinka haga ee la xidhiidhaCiqaabta Ku-noqoshada iyo Koontaroolada Go'aaminta
Su'aal 1 ee 8

Muxuu kor u qaadida cabbirka 'heerkulka' ka qabtaa wax soo saarka moodeelka?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Muxuu kor u qaadida cabbirka 'heerkulka' ka qabtaa wax soo saarka moodeelka?
    • Ka dhigaya mid aan kala sooc lahayn oo kala duwan
    • Waxay ka dhigtaa mid go'aan qaadasho leh oo diiradda saaraya
    • Si joogto ah u beddela miisaanka moodeelka
    • Dadajisa jiilka
  2. Sidee buu shaybaarka top-p (nucleus) u go'aamiyaa calaamadaha la tixgelinayo?
    • Waxay haysaa tiro go'an oo calaamado ah iyadoon loo eegin itimaalka
    • Waxay haysaa calaamada ugu yar kuwaas oo ixtimaalka ay leeyihiin ilaa xad
    • Waxay haysaa oo keliya calaamadda ugu macquulsan
    • Waxay ku haysaa calaamado leh ixtimaalka ka sarreeya 50%
  3. Marxaladee ayuu caadi ahaan ku dhaqmaa ciqaabta ku celiska ah?
    • On qoraalka cayriin jiilka ka dambeeya
    • Ku saabsan xogta tababarka
    • On logits ka hor softmax
    • Ereyada tokenizer-ka
  4. In OpenAI’s text-generation API, how do presence and frequency penalties differ?
    • Joogitaanka miisaanka ciqaabta oo leh tirinta calaamadda; ciqaabta soo noqnoqda waa siman
    • Waa isku mid
    • Labaduba waxay khuseeyaan calaamada koowaad
    • Frequency penalty scales with how often a token appeared; presence penalty applies once if it appeared at all
  5. For code completion where deterministic sampling is desired, which setting best matches the guide?
    • Heerkulka eber u dhow
    • Heerkul aad u sarreeya
    • Ganaaxa soo noqnoqda ee sarreeya ee ka sarreeya 2.0
    • Top-p ee 1.0 oo leh heerkul sare
  6. Muxuu sameeyaa muunada top-k?
    • Waxay haysaa calaamado isku-ururin ilaa ixtimaalka k
    • Waxay u dhigtaa kaliya calaamadaha suurtogalka ah ee k musharax ahaan
    • Ku dhufashada dhammaan logisyada k
    • Waxay ka saartaa calaamadaha ugu yar ee tababarka
  7. With a repetition penalty above 1, how does a standard logits processor adjust a previously seen token?
    • It sets every repeated-token logit to zero
    • It adds the penalty to every token equally
    • It divides a positive logit by the penalty and multiplies a negative logit by it
    • It removes the token from the vocabulary permanently
  8. How does top-k differ from top-p sampling?
    • Top-k uses cumulative probability while top-p fixes a count
    • Top-k changes model weights while top-p changes the tokenizer
    • They are both deterministic decoding methods
    • Top-k keeps a fixed number of candidates; top-p keeps a probability-mass-based set