Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Adapter Layers fun Transfer Quiz

Check how bottleneck adapters preserve a frozen base model, enable modular task or domain transfer, and compare with LoRA.

Jẹmọ itọsọna awọn ọnaAdapter Layers Fun Gbigbe
Ibeere 1 ti 8

Kini yoo ṣẹlẹ si awọn iwọn iṣaju iṣaju atilẹba nigbati ikẹkọ pẹlu awọn oluyipada?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini yoo ṣẹlẹ si awọn iwọn iṣaju iṣaju atilẹba nigbati ikẹkọ pẹlu awọn oluyipada?
    • Wọn ti wa ni aotoju lakoko ti awọn paramita ohun ti nmu badọgba nikan ni ikẹkọ
    • Wọn ti paarẹ
    • Wọn ti tun bẹrẹ laileto
    • Wọn ti wa ni ilọpo meji ni iwọn
  2. Ohun ti o jẹ aṣoju ti abẹnu be ti a bottleneck ohun ti nmu badọgba?
    • A nikan tobi ipon Layer
    • Isọtẹlẹ-isalẹ, aiṣedeede, lẹhinna iṣẹ akanṣe-oke, pẹlu asopọ to ku
    • Nikan ohun akiyesi ori
    • A convolutional image àlẹmọ
  3. Ni aijọju kini ida ti awọn paramita ṣe awọn oluyipada deede ṣe ikẹkọ?
    • 100%
    • Gangan 50%
    • Nigbagbogbo labẹ 5%
    • Ko si rara
  4. Kini idi ti awọn oluyipada jẹ rọrun fun sisin ọpọlọpọ awọn iṣẹ ṣiṣe?
    • Wọn nilo ẹda awoṣe kikun lọtọ fun iṣẹ-ṣiṣe kan
    • Wọn ko le ṣe idapo
    • Wọn nu awoṣe ipilẹ lẹhin iṣẹ-ṣiṣe kọọkan
    • O tọju awoṣe ipilẹ kan ati ki o paarọ kekere awọn alamuuṣẹ iṣẹ-ṣiṣe kọọkan
  5. Si eyi ti ebi ti imuposi ni awọn alamuuṣẹ?
    • Iṣatunṣe-daradara paramita (PEFT)
    • Augmentation data
    • Aworan funmorawon
    • Awọn eto isamisi
  6. Anfani wo ni LoRA funni lori ohun ti nmu badọgba ti a fi sii Ayebaye ni akoko itọkasi?
    • Ko le ṣe ikẹkọ
    • Awọn iwuwo ti o kọ ẹkọ le ṣe idapọpọ si awọn matiriki ipilẹ, yago fun afikun lairi fun-Layer
    • O nilo atunṣe gbogbo awoṣe
    • O yọ asopọ iyokù kuro patapata
  7. What does a residual connection contribute to a classic bottleneck adapter?
    • A second full pretrained model copy for each task
    • A mandatory external retrieval database
    • A path that adds the adapter’s transformed representation back to the original hidden state
    • A guarantee that no added computation occurs
  8. What is one benefit of stacking a domain adapter with a task adapter?
    • Merge all user data into the frozen base weights automatically
    • Avoid using any adapter modules
    • Convert a language model into a vision encoder
    • Reuse a domain specialization together with a separate task specialization