Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele
Nucleus ati Top-k Iṣapeye adanwo
Understand how top-k, top-p and temperature alter token sampling, and why decoding choices change generated-text behavior.
Ibeere 1 ti 8
Bawo ni iṣapẹẹrẹ oke-k ṣe ihamọ awọn yiyan àmi?
Awotẹlẹ gbogbo ibeere8 ibeere
Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.
- Bawo ni iṣapẹẹrẹ oke-k ṣe ihamọ awọn yiyan àmi?
- O tọju awọn ami-iṣeeṣe-k ga nikan, lẹhinna awọn apẹẹrẹ laarin wọn
- O tọju awọn ami titi ti iṣeeṣe akopọ ti kọja k
- Nigbagbogbo o yan aami kan ti o ṣeeṣe julọ
- O yọ awọn k kere seese àmi nikan
- Kini asọye 'nucleus' ninu iṣapẹẹrẹ arin (oke-p)?
- Awọn ọrọ loorekoore k ni data ikẹkọ
- Awọn kere ṣeto ti àmi ti awọn iṣeeṣe apao si ni o kere p
- Awọn nikan ga-iṣeeṣe àmi
- Gbogbo awọn ami pẹlu iṣeeṣe loke 50%
- Kini anfani bọtini ti oke-p lori oke-k?
- O yara nigbagbogbo lati ṣe iṣiro
- O ko gbe awọn ID o wu
- Nọmba awọn ami oludije ṣe deede si igbẹkẹle awoṣe
- O foju pa awọn iṣeeṣe àmi
- Kini paramita iwọn otutu ṣe ṣaaju iṣapẹẹrẹ?
- O ṣeto ipari ti o pọju ọkọọkan
- O yan ede ti iṣelọpọ
- O ka awọn fokabulari iwọn
- O ṣe atunṣe pinpin iṣeeṣe lati jẹ didasilẹ tabi fifẹ
- What result did the nucleus-sampling paper report for its tested open-ended generation tasks?
- It improved diversity without sacrificing fluency and coherence in the reported comparisons
- It guaranteed factual correctness in every output
- It eliminated all sampling randomness
- It improved translation quality on every benchmark
- For a greedy decoding baseline with no token sampling, how should generation be configured?
- Disable sampling and use greedy decoding to select the highest-probability token at each step
- Use high temperature and a large nucleus threshold
- Randomize the temperature at every step
- Keep sampling enabled but increase top-p
- After top-k truncation removes tokens, what happens to the remaining sampling probabilities?
- They remain assigned their original total mass below one
- They are replaced with equal probabilities
- They are renormalized over the retained top-k tokens
- They are converted into beam-search scores
- What did Holtzman et al. report about likelihood-maximizing decoding for open-ended text?
- It always yields the most diverse human-like passages
- O ṣe imukuro iwulo fun awoṣe ede
- It guarantees factual correctness
- It can produce bland and repetitive text despite strong likelihood scores