quiz bu lëkkale ak tegtal · Dëgër Niveau

Decodage spekulatif ak quiz EAGLE

Saytu EAGLE man-manu jëmmal, xool mébet, sàmm séddale, feedback embeddings, garabi jëmmal dinaamik ak njëgu dekode.

Yooni njiit yu jëm ciDecodage spekulatif ak céeli
Laaj 1 de 8

Lan mooy xalaat bu mag bi ci ginaaw dekodage spekulatif?

Xoolaatal laaj bu nekk8 laaj

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

  1. Lan mooy xalaat bu mag bi ci ginaaw dekodage spekulatif?
    • ab defarkat bu yomb defay propoze ay jeton yu bari yu model bu mag bi di xool ci benn pass parallèle
    • kompresse diisaayu model bi ba genn-wàllu njub
    • di sànni ay layer ci reso bi ci anam wu mucc
    • Cache bépp frase bu mëna génn ci kanam
  2. Lan moo wuutale EAGLE ak xeeti dekodaas yu njëkk ya?
    • Dina dindi jéego xool bi yépp
    • Dafay tënk ci barabu màndarga yiñ nëbb ci model bi, du directement ci niveau token bi
    • Dafay jëfandikoo xeetu model bu gëna mag bu wuute lool
    • Dafay dox ludul ci defar nataal
  3. Why is speculative decoding considered lossless when implemented with exact verification?
    • Every draft token is accepted unconditionally
    • The drafter always predicts the target’s next token
    • The acceptance and correction procedure preserves the target model’s output distribution
    • The target model is never evaluated
  4. Ban jafe-jafe ci EAGLE's man-man drafting la feedback token bi njëkk di jàppale ci saafara?
    • Xaajale mémoire ci GPU
    • Tokenizer mënul méngoo ci diggante làkk yi
    • Tàngoor wu ëpp ci aparey bi
    • Ñàkka xam/jaxasoo ci wax luy am ci man-man bi ci topp
  5. Lan la EAGLE-2 yokk ci EAGLE bi njëkk?
    • Garab buy soppiku buy méngale limu kàndidaa yi nga wara tànn
    • ab bordure fixe bu am benn token
    • Dindi boppu LM
    • Taggat ci ñaari yoon done kese
  6. Lan moo waral bottleneck autoregressif bi di def LLM bu leer bi di yeex?
    • Royuwaay bi dafay tàggat boppam ginaaw token bu nekk
    • Bépp jeton bu bees dafa wara jàll ci ay miliyaar ciy parametre balaa beneen bi di tàmbali
    • Token yi deñu leen di denc ci disk bi ci diggante jéego yi
    • GPU bi dafa wara tàmbaliwaat token bu nekk
  7. After EAGLE predicts a hidden-state feature, which component turns that feature into candidate token probabilities?
    • The tokenizer’s vocabulary file alone
    • The optimizer’s momentum buffer
    • The target model’s language-model head
    • The KV-cache allocator
  8. What change does EAGLE-3 make to the original feature-drafting constraint?
    • It replaces the target model with a tokenizer
    • It disables target verification
    • It requires a separate human drafter
    • It removes a feature-prediction constraint to improve scaling