Ikibazo gihuza ibibazo · Biragoye Urwego

Gukora Ibikorwa no Guhagararira Ubwubatsi Ikibazo

Kugenzura uburyo ibikorwa byo kugenzura ibikorwa byagereranijwe, bigashyirwa mu bikorwa kandi bigapimwa, hamwe n'ingaruka ziterwa na rusange.

Ikibazo 1 Bya 8

Nihe gihe mubikorwa byicyitegererezo kuyobora kuyobora bigira uruhare?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Nihe gihe mubikorwa byicyitegererezo kuyobora kuyobora bigira uruhare?
    • Mugihe cyo gufata umwanzuro, muguhindura ibikorwa byihishe
    • Mugihe cyo kwitegura gusa
    • Mugihe cyo gukusanya amakuru
    • Gusa nyuma yo gusohora inyandiko
  2. Nigute icyerekezo kiyobora gikunze kubarwa?
    • Mugutangiza matrise
    • Nka itandukaniro rinini hagati yibikorwa byo gutandukanya ibyiza nibibi
    • Mugusiba ibice kuri moderi
    • Kubara ibimenyetso mubisubizo
  3. Ni ubuhe butumwa bushingiye ku mpamvu yo kuyobora ibikorwa?
    • Igitekerezo cyerekana icyitegererezo kibika amakuru nkumwandiko usobanutse
    • Kwizera ko neuron zose zisa
    • Umurongo ugereranya hypothesis, ko ibintu byinshi ari umurongo ugana icyerekezo
    • Kuvuga ko icyitegererezo kidashobora kwerekana ibitekerezo
  4. Niki coefficente yo gupima kuri vectori igenzura?
    • Ingano yamagambo
    • Umubare wibice muri moderi
    • Igipimo cyo kwiga
    • Ukuntu imyitwarire isunikwa cyane muburyo bwatoranijwe
  5. Ni izihe ngaruka zizwi zo kuyobora moderi bikabije?
    • Ibisohoka hamwe nubushobozi birashobora gutesha agaciro
    • Icyitegererezo cyitoza mu buryo bwikora
    • Ingano ya dosiye ikura bidasubirwaho
    • Yandika burundu uburemere
  6. Nubuhe buryo bukunze guhuzwa nubuyobozi kugirango ubone ibintu bisobanurwa kugirango ufate?
    • Kwiyunvira kwa JPEG
    • Ntibisanzwe autoencoders ibora ibikorwa mubikorwa
    • Kwiyandikisha
    • Disragmentation
  7. How does activation steering differ from fine-tuning the model weights?
    • It retrains the model on a new labeled dataset
    • It changes the tokenizer vocabulary
    • It adds a vector to hidden activations during inference without updating the base weights
    • It replaces the attention layers with a classifier
  8. A steering vector works on the prompts used to compute it but not on a different topic. Which limitation does this illustrate?
    • The vector must be an optimizer state
    • Steering always requires fine-tuning
    • A residual stream contains no features
    • The direction found from one prompt set may not generalize to other contexts