Idanwo ti o sopọ mọ itọsọna · Lile Ipele

PagedAttention ati vLLM adanwo

Test PagedAttention KV blocks, logical-to-physical mapping, and sharing, including the distinction between cache memory savings and total model storage.

Jẹmọ itọsọna awọn ọnaPagedatention Ati Vllm
Ibeere 1 ti 8

Kini 'kaṣe KV' n tọju lakoko iran ọrọ?

Awotẹlẹ gbogbo ibeere8 ibeere

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

  1. Kini 'kaṣe KV' n tọju lakoko iran ọrọ?
    • Awọn data ikẹkọ awoṣe
    • Awọn iwe-ẹri iwọle olumulo
    • Bọtini ati awọn olutọpa iye fun awọn ami ti o kọja ki awọn ami tuntun le lọ si wọn
    • A fisinuirindigbindigbin daakọ ti awọn iwọn awoṣe
  2. Agbekale-eto ẹrọ wo ni atilẹyin PagedAttention?
    • Disk defragmentation irinṣẹ
    • Foju iranti paging
    • Faili funmorawon
    • Sipiyu ẹka asọtẹlẹ
  3. Iṣoro wo pẹlu ipin kaṣe KV ibile ṣe PagedAttention yanju?
    • Awọn isopọ Ayelujara ti o lọra
    • Aini ti ikẹkọ data
    • Aabo ọrọ igbaniwọle ti ko lagbara
    • Egbin iranti lati ifipamọ awọn ohun amorindun nla fun ibeere kọọkan
  4. Bawo ni PagedAttention ṣe jẹ ki awọn abajade lọpọlọpọ pin iranti fun itọsi ti o wọpọ?
    • Awọn ọna oriṣiriṣi tọka si awọn oju-iwe ti ara kanna nipasẹ daakọ-lori-kọ
    • O paarẹ itọka naa lẹhin lilo akọkọ
    • O tọjú awọn tọ lori disk
    • O retrains awọn awoṣe fun kọọkan o wu
  5. Nibo ni vLLM ati PagedAttention ti ni idagbasoke ni akọkọ?
    • A 1990s omowe lab
    • UC Berkeley (ti a tẹjade ni ọdun 2023)
    • A hardware ataja ká tita egbe
    • An Anonymous online forum
  6. Kini anfani iwulo akọkọ ti vLLM fun ẹnikan ti n ṣiṣẹ LLM kan?
    • Ti o tobi awoṣe išedede lori awọn aṣepari
    • Iforukọsilẹ data aifọwọyi
    • Ohun elo GPU ọfẹ
    • Iwajade ti o ga julọ ati awọn ibeere igbakọọkan diẹ sii fun GPU
  7. A sequence’s logical KV block 3 resides in a different physical location. What tells PagedAttention where to read it?
    • The sequence’s block table mapping logical blocks to physical blocks
    • The model’s class-label list
    • The learning-rate scheduler
    • A requirement that every block follow block 2 contiguously
  8. A PagedAttention cache ends with a half-filled block. Which statement about memory waste is accurate?
    • Paging guarantees every allocated byte contains a token
    • The entire model must be copied for each unused slot
    • Unused slots can remain in that final block despite reduced allocation fragmentation
    • The block table cannot represent a partial sequence