quiz bu lëkkale ak tegtal · Diggante Niveau

Guddaay normalisasioŋ ci Quiz biy gëna xéewale tànneef

Saytu ni mébetu tànneef-tuning di doxalee ci guddaayi tontu, lu tax SimPO di normalise poñ token yi, ak ni ñuy jàngatee tontu yu gàtt yi.

Yooni njiit yu jëm ciGuddaay normalisasioŋ ci taamu gëna xéewale
Laaj 1 de 8

Ban jeffin ju baaxul la normalisasioŋ guddaay ci DPO di njëkka jéema tere?

Xoolaatal laaj bu nekk8 laaj

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

  1. Ban jeffin ju baaxul la normalisasioŋ guddaay ci DPO di njëkka jéema tere?
    • Modèle yi bëgg tontu yu gëna gudd, kalite bi du ci dara
    • Model yi nangu wuñu tontu ay laaj
    • Modèle yuy xalaat lu jaarul yoon
    • Model yi tontu ndànk lool
  2. Lan moo waral neexal biñ jagleel DPO dafay gëna yokk lu tontu bi di gëna gudd?
    • Dafay yokk neexal bi ci tàngoor wi
    • Dafay boole wuute yi am ci token bu nekk
    • Dafay jëfandikoo ab bonus bu takku ci frase bu nekk
    • Dafay yar tontu yu gàtt
  3. naka la ko SimPO di def ngir saafara jafe-jafe yi ci guddaay bi?
    • Dafay yokk ñaari yoon diisaayu xeetu royuwaay bi
    • Dafay dagg tontu yépp ci guddaay buñ tëral
    • Dafay jëfandikoo log-probabilite bu token bu nekk ngir neexal
    • Dina dindi done yiñ taamu yépp
  4. Seen wuutu bu am solo ci diggante SimPO ak DPO buñ miin mooy SimPO:
    • Dafay laaj done yu bari ñaari yoon
    • Ci nataal kese lay dox
    • Desactiver wàccinu gradient
    • Jëfandikuwul beneen xeetu royuwaay
  5. Ban jëfu jàngat mooy jàppale ñu firndeel xeetu model bu gëna baax ci kalite te baña yam ci guddaay kese?
    • Raporte ab taux de gagné buñu saytu ci guddaay
    • Lim parametre yépp
    • Natt tàngooru GPU
    • Yokk dayo lots bi
  6. Which chat-based benchmarks did the SimPO paper use in its reported evaluations?
    • ImageNet, COCO, and GLUE
    • AlpacaEval 2, MT-Bench, and Arena-Hard
    • LibriSpeech, WER, and BLEU
    • MMLU, GSM8K, and HumanEval only
  7. What does SimPO’s target reward margin encourage between a preferred and rejected response?
    • Identical scores for both responses
    • A longer preferred response regardless of quality
    • A larger score difference in favor of the preferred response
    • A new reference model for every token
  8. What risk can overly aggressive length normalization create?
    • It guarantees all answers have the same factual accuracy
    • It makes the reference model’s vocabulary larger
    • It prevents evaluators from comparing responses
    • It can favor answers that are too terse to include necessary detail