Guide-yakabatana mibvunzo · Zvakaoma Nhanho

Activation Steering uye Representation Engineering Quiz

Inotarisa kuti activation steering vectors inofungidzirwa sei, inoshandiswa pakufungidzira uye yakayerwa, pamwe nenjodzi kubva kune nekukurumidza-inotsamira generalization.

Nzira dzekutungamira dzinoenderanaActivation Steering uye Representation Injiniya
Mubvunzo 1 ye 8

Ndeipi nguva yekushanda kwemuenzaniso iyo activation steering inopindira?

Tarisa mubvunzo wese8 mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

  1. Ndeipi nguva yekushanda kwemuenzaniso iyo activation steering inopindira?
    • Panguva yekufungidzira, nekugadzirisa zvakavanzika activation
    • Panguva yekudzidzira chete
    • Panguva yekuunganidza data
    • Chete mushure mekunge zvinyorwa zvinobuda zvadhindwa
  2. Ko dhiraivha inofambiswa sei inowanzoverengerwa?
    • Nekungotanga matrix
    • Semusiyano wepakati pakati pe activations pakusiyanisa zvakanaka nezvakaipa zvinokurudzira
    • Nekudzima mitsara kubva kumuenzaniso
    • Nekuverenga zviratidzo mukukurumidza
  3. Ndeipi fungidziro iri pasi pekuti nei activation steering inoshanda?
    • Pfungwa yekuti mamodheru anochengeta data semavara akajeka
    • Kutenda kuti ma neurons ese akafanana
    • Iyo mutsara inomiririra hypothesis, kuti akawanda maficha anenge ane mitsetse nzira
    • Chirevo chekuti mamodheru haagone kumiririra pfungwa
  4. Chii chinoita kuyera coefficient pane inotungamira vheta inodzora?
    • Saizi yemazwi emuenzaniso
    • Nhamba yezvikamu mumuenzaniso
    • Chiyero chekudzidzira kudzidzira
    • Maitiro anosundirwa zvakasimba sei munzira yakasarudzwa
  5. Ndeipi njodzi inozivikanwa yekutungamira modhi zvine hukasha?
    • Kubatana kwekubuda uye kugona kunogona kudzikisira
    • Iyo modhi inozvidzidzisa pachayo
    • Iyo saizi yefaira inokura zvisingadzoreki
    • Inonyora zvakare zviremu
  6. Ndeipi nzira inowanzo batanidzwa neyekutungamirira kuti uwane zvinodudzira maficha ekusungirira?
    • JPEG compression
    • Sparse autoencoders iyo inoparadza activation kuita maficha
    • Domain name registration
    • Disk defragmentation
  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