quiz bu lëkkale ak tegtal · Dëgër Niveau

YaRN ak Quiz buy yokk guddaayi contexte

Saytu ni YaRN di yamale ay fréquence RoPE, di denc ay detay yu bari, di eskale tàngoor wi ñuy bàyyi xel, di yaatal contexte bi ak ay fine-tuning.

Yooni njiit yu jëm ciYarn ak yaatuwaayu guddaayi contexte
Laaj 1 de 8

Ban anam enkodaasu position la YaRN di soppi ngir yokk muy wax?

Xoolaatal laaj bu nekk8 laaj

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

  1. Ban anam enkodaasu position la YaRN di soppi ngir yokk muy wax?
    • Rope (RoPE)
    • absolu sinusoïdale
    • Tablo seetlu position yiñ jàng
    • matris bias relatif kese
  2. Lan mooy jaarul yoon sudee modelu RoPE gisna ay toppalante yu gëna gudd guddaayi tàggat yaram?
    • Tokenizer bi dafa tass
    • Wëréelu angle yi dañuy dugg ci diggante yuñu gisul, ba noppi performance yi dañuy daanu
    • Modèle bi amul lenn ludul génne
    • Degrade yi dañuy nekk benn
  3. naka la YaRN di def ay dimension fréquence RoPE yu wuute?
    • Dafay boole dimension yépp ci anam wu tolloo
    • Dafay efaase dimension yu fréquence yu kawe yi
    • Dafay boole dimension yu fréquence yu woyof yi, di denc yu fréquence yu kawe yi
    • Dafay aleatoire bépp dimension
  4. Lu weesu wàññiwaat fréquence yi, ban coppite la YaRN di def?
    • Tokeniseer bu bees
    • Ñaari yoon limu diisaay yi
    • Dindi normalisasioŋ couche
    • Yamale tàngoor (softmax scaling)
  5. Lan mooy njariñu YaRN ci kaw interpolation bu xamul dara?
    • Dafay yegg ci contexte bu gudd ak ay done ak jéego yu néew
    • Soxlawul benn model
    • Du dox ludul ci suufu 512 token
    • Dafay dindi soxla GPU yi
  6. Lan moo waral interpolaasioŋ ligneer bu yomb ci fréquence RoPE yépp luy lore?
    • Dafay indi jafe-jafe ci wàllu parçage
    • Dafay tënk dimension yu am fréquence yu bari yuy kode ay detay yu am ci gox bi
    • Dafay gëna rëy model bi
    • Dafay soppi dayo vokabileer bi
  7. In RoPE, which vectors receive position-dependent rotations?
    • Only the value vectors
    • Only the model’s token IDs
    • The query and key vectors
    • The optimizer’s gradient and momentum
  8. What does YaRN’s attention-temperature adjustment address when context length is extended?
    • The number of GPU layers
    • The dimensions of the token vocabulary
    • The weight decay coefficient
    • Changes in attention entropy or softmax sharpness caused by the longer context