Guide-yakabatana mibvunzo · Pakati Nhanho

Logit Lens uye Tuned Lens Quiz

Inotarisa zvepakati-layer kuverenga, yakasara-rukova chimiro, iyo yakagadziridzwa lens muturikiri uye zvipimo zveraw logit lens.

Nzira dzekutungamira dzinoenderanaLogit Lens uye Tuned Lens
Mubvunzo 1 ye 8

Logit lens inoita sei?

Tarisa mubvunzo wese8 mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

  1. Logit lens inoita sei?
    • Inoshandisa iyo yekupedzisira kusamira kweiyo modhi kune yepakati yakavanzika nyika kuverenga fungidziro pachikamu chimwe nechimwe
    • Inowedzera mitsara mitsva kune transformer
    • Inobvisa kutarisa kubva kumuenzaniso
    • Inoturikira pakati pemitauro yevanhu
  2. Ndeipi kuvandudza kwakakosha iyo lens yakaturikirwa inowedzera pamusoro peiyo mbishi logit lens?
    • Inoshandisa izwi guru
    • Inodzidzisa diki per-layer affine mepu kushandura nyika dzakavanzwa kuita hwaro hwekupedzisira
    • Inobvisa kudiwa kwechero kusaina
    • Inoshanda chete pamifananidzo yemifananidzo
  3. Ndechipi chimiro mumatransformer chinoita kuti lenzi idzi dzigoneke?
    • The convolutional filters
    • The tokenizer
    • Iyo yekuwedzera yasara rwizi iyo iyo isina kuvharwa gare gare inogadziridza kune marogi
    • The optimizer's kudzidza mwero
  4. Ndiani akaunza yekutanga logit lens, uye riini?
    • DeepMind muna 2015
    • OpenAI muna 2018
    • Google muna 2012
    • nostalgebraist muna 2020
  5. Ndekupi iyo logit lens inowanzoratidza mhinduro yekupedzisira 'crystallizing'?
    • Pakati pepakati-kusvika-kunonoka
    • Muchikamu chekutanga chete
    • Kamwe, mhinduro inooneka yese kamwechete
    • Kunze kwemuenzaniso zvachose
  6. Nei iyo mbishi logit lens ingave isingavimbike mumatanho ekutanga?
    • Matunhu ekutanga haana tsinga
    • Mamiriro avo akavanzika haasati aenderana neyokupedzisira kuburitsa hwaro iyo inomisikidzwa inotarisirwa
    • Mavara ekutanga haamboshandiswa mukufanotaura
    • Iyo unembedding matrix inongoitika
  7. What does the model’s final unembedding map a residual-stream vector into?
    • A new set of transformer layers
    • A similarity graph over training examples
    • Vocabulary logits that can be converted into token probabilities
    • A gradient clipping threshold
  8. Why can an early-layer raw logit-lens prediction be misleading?
    • The layer has already generated the full answer
    • The tokenizer has been removed
    • The unembedding only works on images
    • The intermediate state may not yet be aligned with the final output basis