Guide-yakabatana mibvunzo · Pakati Nhanho

KV Cache Quiz

Ongorora KV-cache kushandiswa zvakare panguva yekuzvimiririra decoding, ndangariro kumanikidzwa, akawanda-chikumbiro kugovera uye mapeji ekugovera tradeoffs.

Nzira dzekutungamira dzinoenderanaKv Cache
Mubvunzo 1 ye 8

Ko KV cache inochengeterei?

Tarisa mubvunzo wese8 mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

  1. Ko KV cache inochengeterei?
    • Makiyi uye kukosha mavector akaverengerwa kune apfuura tokens
    • Iyo modhi izere huremu matrices
    • Chat nhoroondo yemushandisi semavara akajeka
    • Mhedzisiro inobuda
  2. Nderipi dambudziko iro KV cache inonyanya kugadzirisa?
    • Inovandudza girama yezvinyorwa zvakagadzirwa
    • Iyo inodzivirira kudzoreredza makiyi uye kukosha kune ese epamberi tokens padanho rega rega
    • Inowedzera mazwi emuenzaniso
    • Inonyora nhaurirano
  3. Chii chakanyanya kuderera cheKV cache?
    • Zvinoita kuti chizvarwa chinonoke
    • Inoderedza kubuda kwehutano
    • Kushandiswa kwendangariro kwayo kunokura zvakatevedzana nehurefu hwechirevo
    • Inoshanda chete paCPUs
  4. Ndeipi nzira inoderedza saizi yeKV cache nekugovana makiyi uye kukosha pamisoro yekutarisisa?
    • Dropout
    • Beam search
    • Kukamura Mazwi
    • Mumapoka-mubvunzo kana mizhinji-mubvunzo kutarisa
  5. Ndeapi zvikamu zviviri zveshanduro yetransformer inoenderana neKV cache?
    • Prefill (gadzirisa kukurumidza) uye decode (gadzira chiratidzo chimwe panguva)
    • Kudzidziswa uye kuedza
    • Encode uye compress
    • Isa uye dhawunirodha
  6. Chii chinoitwa nevLLM's PagedAttention kune KV cache?
    • Inodzima cache mushure mechiratidzo chega chega
    • Inogovera iyo cache mumabhuraki asina- contiguous kuderedza kupatsanurwa kwendangariro
    • Inoshandura cache kuita zvinyorwa
    • Inopeta kaviri nhamba yeparamende yemuenzaniso
  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