Ikibazo gihuza ibibazo · Hagati Urwego

Gusubiramo Ibihano no Kugenzura Igenzura

Gereranya ubushyuhe, top-k, top-p, nibihano byo gusubiramo, harimo nuburyo logi ihinduka mbere yo gutoranya mubikorwa bya LLM.

Inzira zijyanye no kuyoboraGusubiramo Ibihano no Kugenzura Igenzura
Ikibazo 1 Bya 8

Niki kuzamura ibipimo bya 'ubushyuhe' bikora kubisubizo by'icyitegererezo?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Niki kuzamura ibipimo bya 'ubushyuhe' bikora kubisubizo by'icyitegererezo?
    • Bituma bidashoboka kandi bitandukanye
    • Bituma irushaho gushishoza no kwibanda
    • Guhindura burundu uburemere bwikitegererezo
    • Kwihutisha ibisekuruza
  2. Nigute top-p (nucleus) icyitegererezo cyerekana ibimenyetso ugomba gusuzuma?
    • Igumana umubare uhamye wibimenyetso utitaye kubishoboka
    • Igumana utuntu duto duto twerekana ibimenyetso bishoboka kurwego
    • Igumana gusa ikimenyetso kimwe gishoboka
    • Ikomeza ibimenyetso hamwe nibishoboka hejuru ya 50%
  3. Ni ikihe cyiciro igihano cyo gusubiramo gikora?
    • Ku nyandiko mbisi nyuma yisekuru
    • Ku makuru y'amahugurwa
    • Kuri logi mbere ya softmax
    • Kuri tokenizer amagambo
  4. In OpenAI’s text-generation API, how do presence and frequency penalties differ?
    • Kubaho umunzani wibihano hamwe no kubara ibimenyetso; igihano cyinshyi kiringaniye
    • Birasa
    • Byombi bikurikizwa gusa kubimenyetso byambere
    • 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?
    • Ubushyuhe hafi ya zeru
    • Ubushyuhe bwo hejuru cyane
    • Igihano cyo gusubiramo hejuru hejuru ya 2.0
    • Hejuru-p ya 1.0 hamwe n'ubushyuhe bwo hejuru
  6. Gukora top-k gukora iki?
    • Komeza ibimenyetso byerekana incamake k
    • Gumana gusa k ibishoboka bishoboka nkabakandida
    • Kugwiza inyandiko zose kuri k
    • Kuraho k byibuze ibimenyetso byerekana mumahugurwa
  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