Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Agbelebu-Akiyesi adanwo

Understand how cross-attention lets one sequence or modality query representations from another sequence or modality in transformer models.

Jẹmọ itọsọna awọn ọnaAgbelebu Ifarabalẹ
Ibeere 1 ti 8

Kini iyatọ bọtini laarin akiyesi ara ẹni ati akiyesi agbelebu?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini iyatọ bọtini laarin akiyesi ara ẹni ati akiyesi agbelebu?
    • Ni akiyesi agbelebu, awọn ibeere wa lati ọna kan ati awọn bọtini / iye lati ọdọ miiran
    • Cross-akiyesi nlo ko si softmax
    • Ifarabalẹ ti ara ẹni ṣiṣẹ nikan lori awọn aworan
    • Ifarabalẹ-agbelebu ko le ṣe ikẹkọ pẹlu isunmọ gradient
  2. Ninu Oluyipada koodu koodu, nibo ni awọn ibeere fun akiyesi agbelebu ti wa?
    • Awọn abajade kooduopo
    • Ipo decoder lọwọlọwọ
    • Awọn ifibọ aise igbewọle nikan
    • A lọtọ ID matrix
  3. Kilode ti awọn ọna meji ti o wa ni ifarabalẹ-agbelebu ni awọn ipari gigun tabi awọn ipo?
    • Nitoripe akiyesi nilo awọn igbewọle gigun-dogba
    • Nitoripe o kọju ọkan ninu awọn ilana naa patapata
    • Nitoripe awọn ibeere ati awọn bọtini wa lati oriṣiriṣi, awọn orisun ominira
    • Nitoripe o pa ohun gbogbo lọ si iwọn ti o wa titi ni akọkọ
  4. Bawo ni Itankale Stable ṣe lo akiyesi agbelebu?
    • Lati compress awọn aworan sinu awọn faili kekere
    • Lati tumọ ọrọ laarin awọn ede
    • Lati yọ ariwo kuro ninu ohun
    • Lati ipo awọn agbegbe aworan ti ipilẹṣẹ lori ọrọ tọ
  5. Iṣẹ ṣiṣe iṣiro wo ni akiyesi agbelebu pin pẹlu akiyesi ara ẹni?
    • Ifojusi aami-ọja ti iwọn pẹlu softmax kan
    • Iyika lori awọn piksẹli
    • Loorekoore ipinle awọn imudojuiwọn
    • Awọn tabili wiwa koodu lile
  6. What gate does Flamingo use to control information flow through its cross-attention layers?
    • A hard threshold initialized to one
    • A tanh gate initialized to zero
    • A fixed residual weight of 0.5
    • A learned vocabulary lookup table
  7. What input sequence structure does Flamingo’s paper say its visual-language model can handle?
    • Only one image followed by a fixed-length label
    • Only text tokens with no visual inputs
    • Visual and textual data interleaved in arbitrary sequences
    • Only video frames with no text
  8. What is one purpose of cross-attention in Flamingo’s visual-language architecture?
    • It replaces all text tokens with pixels
    • It prevents any information from passing between modalities
    • It performs only image resizing
    • It lets language-model layers incorporate visual features from image/video inputs