Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Kediska KV Cache

Dib u eeg dib-u-isticmaalka KV-cache inta lagu guda jiro goynta autoregressive, cadaadiska xusuusta, wadaaga codsiyada badan iyo ganacsiga qoondaynta bogga.

Wadooyinka haga ee la xidhiidhaKaydka Kv
Su'aal 1 ee 8

Muxuu kaydiyaa kaydka KV?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Muxuu kaydiyaa kaydka KV?
    • Furaha iyo qiyamka vectors ayaa lagu xisaabiyay calaamadihii hore
    • Miisaanka miisaankiisa buuxa ayaa isbarbar dhigaya
    • Taariikhda sheekeysiga isticmaalaha sida qoraal cad
    • Itimaalka wax soo saarka kama dambaysta ah
  2. Dhibaato noocee ah ayuu ugu horrayn xalliyaa kaydka KV?
    • Waxay wanaajisaa naxwaha qoraalka la sameeyay
    • Waxay ka fogaataa dib u xisaabinta furayaasha iyo qiyamka dhammaan calaamadihii hore ee tallaabo kasta
    • Waxay kordhisaa ereyada moodeelka
    • Way siraysaa wada hadalka
  3. Waa maxay cilladda ugu weyn ee kaydka KV?
    • Waxay ka dhigtaa jiilka mid gaabis ah
    • Waxay yaraynaysaa tayada wax soo saarka
    • Isticmaalkeedu wuxuu u koraa si toos ah oo leh dhererka macnaha guud
    • Waxay kaliya ku shaqeysaa CPU-yada
  4. Farsamokee ayaa yaraynaysa cabbirka kaydka KV iyadoo la wadaagayo furayaasha iyo qiyamka madaxa dareenka?
    • Ka saarida
    • Baadhitaanka Beam
    • Calaamadaynta
    • Su'aal kooxaysan ama feejignaan su'aalo badan
  5. Waa maxay labada weji ee soo jeedinta transformer marka loo eego kaydka KV?
    • Horu-buuxinta (ka socodsii dagdagga) oo kala saar (halkii hal calaamad u samee)
    • Tababar iyo imtixaan
    • Codsi oo cadaadi
    • Soo rar oo soo deji
  6. Muxuu vLLM's PagedAttention ka qabtaa kaydka KV?
    • Waxay tirtirtaa kaydka calaamad kasta ka dib
    • Waxay u qoondaysaa kaydka meelo aan xidhiidhsanayn si loo yareeyo kala qaybsanaanta xusuusta
    • U beddelo kaydka qoraal
    • Waxay labanlaabaysaa tirinta cabbirka moodeelka
  7. Why can KV-cache memory limit how many requests a GPU serves concurrently?
    • The cache stores all training examples on the GPU
    • Each request creates a new model copy
    • Each active sequence retains key and value tensors that grow with its generated context
    • The cache stores only final text on disk
  8. What is one benefit of sharing KV-cache blocks among requests with a common prefix?
    • It makes unrelated prefixes identical
    • It removes the need to compute attention
    • It guarantees lower latency in every workload
    • It can avoid storing duplicate prefix state and reduce memory use