njaxasu laaj ak tontu bu kàngam yi ci LoRA

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

1. Lan mooy wàllu 'LoRA' ci njaxasu kàngam yi ci LoRA?
   • Adaptaasioŋ bu am rang bu woyof, di tàggat matris yu ndaw yu am rang bu woyof ci barabu poid yu mat
   • Royuwaayu neexal jàng buy dooleel
   • Dañuy bàyyi xel ci anam wu sori
   • Tuyo buy yab ay done ngir adaptatër yi
2. Lan mooy liggéeyu reso gating/routeur ci MoLE?
   • ngir kompresse diisaayu modelu base
   • Xalaatal ban kàngam ngay aktive ak ban poid ngir benn dugal
   • Defar ay done tàggat yaram ngir adaptër yi
   • Ngir soppi mbind mu nekk ay samp
3. Lan moo waral MoLE di gëna baax tàggat benn adaptatër ci liggéey yu bari te wuute?
   • Dafay jëfandikoo mémoire GPU bu gëna bari ci jëmmal
   • Dafay dindi soxla benn modelu base
   • Dafay wàññi jaxasoo ak fàtte ci denc xam-xam ci ay kàngam yuñ mëna boole
   • Benn làkk kese lay dox
4. Ci LoRA, yeesali diisaay bi, ban ci yii lañuy faral di fësal?
   • matrix bu dëgër buy wecci W
   • Benn xeetu eskalaar ci kaw W
   • Jéego bu nekk dañuy yokk benn matrix bu bari bruit
   • Produit bu ñaari matris yu rang bu woyof, B yoon A
5. Bu ñuy tàggat MoLE, lan mooy xew ci diisaay yiñ tàggat bu njëkk ci modelu base bi?
   • Dañuy des ci gelé bi adaptatër yi ak routeur bi kese lañuy tàggat
   • Dañu leen di tàggataat ci noonu rek
   • Dañu leen di efaase ba noppi tabaxaat leen
   • Dañu leen di xayma ci benn bit
6. Ci misaali nataali diffusion, ban jafe-jafe la gating hierarchique/per-couche ci MoLE di jàppale ci saafara?
   • Yebbi internet bu yeex
   • konsept LoRA yu bari di daanele seen biir suñu leen boole
   • Ñàkum done ci wàllu tàggat yaram
   • Format fichier yu jaarul yoon
7. How can a gated LoRA mixture differ from merging all adapter weights into one fixed update?
   • It must retrain the pretrained backbone for every prompt
   • It can use only one expert for all inputs
   • It can select or weight experts conditionally for a task or layer
   • It removes every low-rank matrix before inference
8. In a LoRA update with rank r, what does r control most directly?
   • The number of decoder tokens generated
   • The number of original model layers removed
   • The size of the training vocabulary
   • The dimension of the low-rank factors and thus the capacity and parameter cost of the update