Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Kediska Saadaasha Token Xiga

Dib u eegis tabobarka qaabka luuqada isdifaacida, macnaha guud, ku qasbaya macalinka iyo faraqa u dhexeeya tababarka iyo jiilka.

Wadooyinka haga ee la xidhiidhaNext Token Saadaasha
Su'aal 1 ee 8

Muxuu soo saarayaa tusaalaha saadaasha xigta ee tallaabo kasta?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Muxuu soo saarayaa tusaalaha saadaasha xigta ee tallaabo kasta?
    • Qaybinta itimaalka ee ku saabsan ereyada calaamada xigta
    • Hal kelmad go'an
    • Turjumaadda jumlada
    • Dhibcaha naxwaha
  2. Dareenkee noocee ah ayaa isticmaala koodhka qaabka GPT?
    • Fiiro gaar ah labada dhinac
    • Sababta (masgaran) dareenka qarinaysa calaamadaha mustaqbalka
    • Wax dareen ah haba yaraatee
    • Fiiro gaar ah
  3. Waa maxay macnaha jiilka 'autoregressive'?
    • Qaabku wuxuu soo saaraa dhammaan calaamadaha hal mar
    • Qaabku wuxuu dib ugu noqdaa qiime dhexdhexaad ah
    • Calaamad kasta oo cusub waa lagu lifaaqay oo dib ayaa loo soo celinayaa si loo saadaaliyo kan xiga
    • Calaamadaha waxaa loo soo saaray si aan kala sooc lahayn
  4. Waa maxay shaqada khasaaraha sida caadiga ah la yareeyey inta lagu jiro tababarka?
    • Khalad afar gees leh celcelis ahaan
    • Luminta hinge
    • Fogaanta cosine
    • Cross-entropy ka sarreeya calaamadda xigta
  5. Maxaa kor u qaadaya heerkulka inta lagu jiro tijaabada?
    • Waxay ka dhigtaa wax soo saarka si aan kala sooc lahayn oo hal abuur leh
    • Waxay ka dhigtaa wax soo saarka mid si adag loo go'aamiyo
    • Waxay kordhisaa cabbirka erayada
    • Waxay ka saartaa lakabka softmax
  6. Ka hor saadaasha, sidee baa qoraalka cayriin loo diyaariyaa?
    • Waxaa loo turjumay Ingiriis
    • Waxa loo kala qaybiyaa calaamado by tokenizer
    • Waxa loo rogaa maqal
    • Waxaa looga tagay sidii jilayaal cayriin ah oo aan lahayn habayn
  7. During next-token training, what is teacher forcing?
    • The model receives future target tokens before the input
    • The model randomly changes the tokenizer at every step
    • The model receives the ground-truth preceding tokens when predicting the next token
    • The model always samples its own previous prediction during training
  8. What mismatch can arise because generation feeds back model-produced tokens?
    • The model can no longer use attention
    • The vocabulary disappears after the first token
    • The model’s loss is computed only on images
    • Errors in generated prefixes can affect later predictions, unlike training on the ground-truth prefix