Next-Tokini Asọtẹlẹ adanwo

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

1. Kini abajade awoṣe asọtẹlẹ ami atẹle ni igbesẹ kọọkan?
   • A iṣeeṣe pinpin lori awọn fokabulari fun awọn tókàn àmi
   • Ọrọ kan ti o wa titi
   • Itumọ gbolohun naa
   • Dimegilio girama kan
2. Iru akiyesi wo ni oluyipada ara GPT nlo?
   • Ifojusi bidirectional
   • Ifarabalẹ ti o fa (boju-boju) ti o tọju awọn ami ami iwaju
   • Ko si akiyesi rara
   • Ifojusi aifọwọyi
3. Kini iran 'autoregressive' tumọ si nibi?
   • Awọn awoṣe gbogbo awọn ami ni ẹẹkan
   • Awọn awoṣe regresses to a tumosi iye
   • Kọọkan titun àmi ti wa ni appended ati ki o je pada ni lati ṣe asọtẹlẹ tókàn
   • Àmi ti wa ni ti ipilẹṣẹ ni ID ibere
4. Iṣẹ ipadanu wo ni a dinku ni igbagbogbo lakoko ikẹkọ?
   • Aṣiṣe onigun mẹrin tumọ si
   • Pipadanu mitari
   • Ijinna kosin
   • Cross-entropy lori tókàn àmi
5. Kini igbega iwọn otutu lakoko iṣapẹẹrẹ ṣe?
   • Mu ki o wu diẹ ID ati ki o Creative
   • Mu ki o wu muna deterministic
   • Ṣe alekun iwọn fokabulari
   • Yọ awọn softmax Layer
6. Ṣaaju asọtẹlẹ, bawo ni a ṣe pese ọrọ aise nigbagbogbo?
   • O tumọ si Gẹẹsi
   • O ti pin si awọn ami-ami nipasẹ tokenizer
   • O ti wa ni iyipada si iwe ohun
   • O ti wa ni osi bi aise ohun kikọ pẹlu ko si processing
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