Ikibazo gihuza ibibazo · Hagati Urwego
Ikibazo n'Ikibonezamvugo Cyayobowe n'Ikibazo
Ongera usuzume amategeko ya decoding yemewe na parser leta, hanyuma usuzume uko ibintu byarangiye nibishobora guhinduka binyuze muri JSON hamwe na mask-sample ingero.
Ikibazo 1 Bya 8
Ni ikihe cyiciro imbogamizi zasobanuwe zisobanura kugabanya imiterere y'ururimi guhitamo?
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Ni ikihe cyiciro imbogamizi zasobanuwe zisobanura kugabanya imiterere y'ururimi guhitamo?
- Only while collecting pretraining documents
- Only after executing every emitted instruction
- Decoding, before the next token is selected
- By retraining the model weights after every user request
- Ni ubuhe buryo bwo guhagararira amategeko buganirwaho mu bushakashatsi bwakozwe mu kibonezamvugo?
- Only image-file color palettes
- Only a spreadsheet of model parameter counts
- Only an alphabetical list of document filenames
- Regular expressions and context-free grammars
- Ni ukubera iki imashini yoroheje ya leta itari ibisobanuro byuzuye bya buri kintu rusange cyashyizwe mu majwi kitarangwamo ikibonezamvugo?
- General nested structure can require stack-like or richer parser state
- All nested grammars must be regular expressions by definition
- A grammar can never describe nesting
- Every parser must discard its current state after each token
- Ni izihe ngorane nyamukuru za tekinike mugushira mubikorwa inzitizi-urwego?
- Icyitegererezo kibura kwibuka ako kanya
- Imipaka ya Tokenizer ntishobora gutondekanya imipaka yikibonezamvugo
- Ikibonezamvugo ntigishobora kwandikwa kuri JSON
- GPUs ntishobora kugwiza matrices
- Ni ubuhe buryo bwo gusaba bufite ishingiro kubisohoka birangira neza munsi yikibonezamvugo gishyigikiwe neza?
- Every fact inside it is necessarily true
- Every possible interruption leaves a complete accepted output
- Every downstream application rule is automatically covered
- It satisfies the constraints covered by that grammar
- Which named research engine studies flexible and efficient structured generation for language models?
- A photo-upscaling file format
- An audio loudness unit
- XGrammar
- A required model-weight encryption algorithm
- Generation stops after only an opening JSON object brace. What can be concluded about that text?
- It must already be a complete JSON object
- It can be a permitted prefix without being a complete valid object
- The interruption proves no JSON grammar exists
- Every such prefix is necessarily forbidden from the start
- A stated distribution assigns A 0.2, B 0.3, and C 0.5. After C is masked and the remaining choices are renormalized, what is A's probability?
- 0.4
- 0.2
- 0.5
- 0.6