Maswali yanayounganishwa na mwongozo · Kati Kiwango

Maswali ya Kubadilisha Miundo

Jaribu majukumu ya vivuta na viboreshaji, linganisha viainishi vya msalaba na viainishi vya mbili, na uchunguze upangaji wa orodha ya seti za watahiniwa.

Njia za mwongozo zinazohusianaKuweka upya Miundo
Swali 1 ya 8

Katika bomba la kawaida la urejeshaji wa hatua mbili, kazi ya mpangaji upya ni nini?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Katika bomba la kawaida la urejeshaji wa hatua mbili, kazi ya mpangaji upya ni nini?
    • Ili kupata tena na kupanga upya orodha fupi ya wagombeaji kwa umuhimu
    • Kuvuta watahiniwa wa awali kutoka mamilioni ya hati
    • Kuhifadhi hati katika hifadhidata
    • Ili kutoa jibu la mwisho la lugha asilia
  2. Besides MS MARCO, on which dataset did Nogueira and Cho report strong results for their BERT passage reranker?
    • Picha ya Picha
    • TREC-CAR
    • LibriSpeech
    • SQuAD only
  3. Kwa nini kiweka upya kisimbaji upya hakiwezi kutumiwa kutafuta mamilioni ya hati moja kwa moja?
    • Inafanya kazi kwenye picha tu
    • Haiwezi kusoma maandishi
    • Alama zake lazima zihesabiwe upya kwa kila jozi ya hati ya hoja, ambayo ni ya polepole sana kwa kiwango.
    • Haihitaji hesabu hata kidogo
  4. Je, ni faida gani kuu ya kisimbaji-mbili juu ya kisimbaji-mkataba?
    • Ni sahihi zaidi katika umuhimu uliowekwa vizuri
    • Daima hupanga upya matokeo kikamilifu
    • Haihitaji data ya mafunzo
    • Upachikaji wa hati zake unaweza kukokotwa mapema, na kufanya urejeshaji haraka na uongezeke
  5. How can a reranker affect context quality in a retrieval-augmented generation pipeline?
    • It can move more relevant retrieved candidates higher in the context shortlist, but does not guarantee correctness
    • It ensures every retrieved passage is factually correct
    • It creates missing documents in the corpus
    • It removes the need to evaluate retrieval
  6. In listwise reranking, what does the model consider together?
    • Only one document with no query
    • A query and a set of candidate documents when deciding their ordering
    • Each document embedding before any candidate retrieval
    • A single relevance label for the full training corpus
  7. What should a team verify before interpreting reranker scores as probabilities?
    • Whether every score is above zero
    • Whether the first-stage retriever was removed
    • Whether the scores have been calibrated and evaluated for that use
    • Whether the model has more parameters than the retriever
  8. What limitation does a reranker inherit from the first-stage retrieval system?
    • It always returns a calibrated probability
    • It removes the need for retrieval
    • It guarantees a correct generated answer
    • It cannot promote a relevant document that was never included in the candidate set