Kedis ku xidhan hagaha · Adag Heerka

Xoojinta Fine-Tuning with Grader Quiz

Wuxuu tijaabiyaa sida ardaydu u dhaliyaan wax soo saarka moodeelka, kuwaas oo hawluhu ku habboon yihiin xoojinta hagaajinta wanaagsan, iyo sida abaalgudka jabsiga iyo calaamadaha tababbarka daciifka ahi u soo baxaan.

Wadooyinka haga ee la xidhiidhaXoojinta Daawaynta Fiican ee Ardayda
Su'aal 1 ee 8

Sidee buu u hagaajinta hagaajinta uga duwan yahay hagaajinta la kormeerayo?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Sidee buu u hagaajinta hagaajinta uga duwan yahay hagaajinta la kormeerayo?
    • It trains the model to imitate exact target outputs
    • It rewards sampled outputs a grader scores highly instead of copying examples
    • It needs no training prompts
    • It only changes the system prompt
  2. Nooca ardaydee ayaad isticmaali lahayd si aad u hubiso in koodka la sameeyay uu run ahaantii shaqeeyo?
    • A programmatic grader that runs unit tests
    • A string-match grader comparing to one reference function
    • A grader that counts words
    • No grader is needed for code
  3. Shaqadee ayaa ugu liidata ku haboon RFT marka loo eego hagaha?
    • Classifying items into defined category codes
    • Solving maths problems with known answers
    • Judging subjective quality where expert graders disagree
    • Structured data extraction with a reference answer
  4. Xoojinta Fine-Tuning with Graders: waa maxay abaalgudka jabsiga?
    • Stealing another team's grader
    • The model learning to satisfy the grader rather than the real intent
    • Using too little compute during training
    • A grader giving partial credit
  5. Maxay tahay sababta keentay in muunad kasta uu helo isla dhibco ay bixiso tilmaamo waxbarasho oo yar oo ah hababka qaraabada ah ee kooxda sida GRPO?
    • Because the grader crashes on them
    • Because learning comes from score differences within the group
    • Because they are too long
    • Because GRPO ignores scores entirely
  6. What is the purpose of a KL penalty against the original model during RFT?
    • To speed up grading
    • To limit how far the model drifts from its starting behaviour
    • To increase output length
    • To replace the grader
  7. A model grader over-rewards confident, lengthy replies. What is the likely outcome of training against it?
    • Outputs drift toward longer, more confident wording
    • Qaabku wuxuu joojiyaa soo saarista
    • Accuracy is guaranteed to improve
    • The grader becomes stricter automatically
  8. Which is the best way to check that rising training reward reflects real improvement?
    • Trust the training reward curve
    • Increase the learning rate
    • Spot-check outputs by hand or with an independent grader on held-out data
    • Remove the validation set