Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Kediska Madaxyada Go'aaminta Medusa

Dib u eeg madaxa saadaasha calaamada badan ee Medusa, geedaha musharaxa, xaqiijinta foojignaanta geed, aqbalida horgalaha ugu dheer, iyo ilaalinta qaybinta.

Wadooyinka haga ee la xidhiidhaMadaxa Goynta Medusa
Su'aal 1 ee 8

Maxay saadaaliyeen madaxda Medusa ee dheeraadka ah?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Maxay saadaaliyeen madaxda Medusa ee dheeraadka ah?
    • Calaamado mustaqbalka oo badan oo meelo kala duwan ka horeeya
    • Luminta tababarka moodeelka
    • Waa kuwee GPU in la isticmaalo
    • Cinwaanka sawirka calaamad
  2. Under which Medusa configuration does the paper describe lossless inference acceleration?
    • Waxay si fiican u cadaadinaysaa miisaanka modelka
    • Medusa-1 with a frozen backbone and exact verification against the original model
    • Marna khalad ma samaynayso naxwaha
    • Waxay meesha ka saaraysaa baahida eray bixinta
  3. Qaab noocee ah ayaa sii wadida musharaxiinta Medusa loo diyaariyey si loo xaqiijiyo?
    • Liis siman ayaa mid mid dhaliyay
    • Miiska xashiishka
    • Geed ayaa dhaliyay hal baas oo leh maaskaro dareenka geedka
    • Wareeg soo noqnoqda oo ka sarreeya calaamad kasta
  4. Waa maxay faa'iidada muhiimka ah ee Medusa marka loo eego qaab-qabyo-qodoynta mala-awaalka ah?
    • Waxay u baahan tahay laba nooc oo waaweyn oo kala duwan
    • Waxay kaliya ku shaqeysaa sawirada
    • Waxay tirtirtaa qaabka aasaasiga ah ka dib isticmaalka
    • Waxay ka fogaanaysaa u baahan qaab qabyo ah oo gaar ah iyadoo ku dartay madax mid jira
  5. Qiyaastii xawaare noocee ah ayay Medusa caadi ahaan ka warbixisaa?
    • Qiyaastii 2-3x
    • Qiyaastii 0.5x (kayar)
    • Dhab ahaan 100x
    • Isbeddel la qiyaasi karo ma jiro
  6. Waa maxay madaxa Medusa shaqsi kasta, qaab dhismeed ahaan?
    • Transformer buuxa oo nuqul ah
    • Lakab yar oo haraaga feed-horumarin ah (MLP) oo ku yaal dusha sare ee gobolka qarsoon
    • Codeeyaha sawirka convolutional
    • Calaamad gaar ah
  7. What does Medusa accept from a verified candidate tree?
    • Every speculative branch regardless of verification
    • Only the shortest candidate token
    • The longest prefix consistent with the base model’s decoding result
    • A random leaf from the tree
  8. How do Medusa’s prediction-head offsets differ?
    • Every head predicts the same next token
    • Each head predicts a different vocabulary
    • The heads predict tokens only after decoding is complete
    • Each head predicts a token at a different position ahead in the sequence