Ijusile Iṣapẹẹrẹ Fine-Tuning adanwo

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

1. Iru lupu wo ni o dara julọ ṣe apejuwe bi Ijusilẹ Iṣayẹwo Fine-Tuning ṣe ilọsiwaju awoṣe?
   • Ṣe ipilẹṣẹ ọpọlọpọ awọn idahun oludije, tọju awọn ti o ni igbelewọn ti o dara julọ, ki o tun ṣe ikẹkọ lori wọn
   • Pa awọn ipele awoṣe ti n ṣiṣẹ buru ju
   • Kọ gbogbo data ikẹkọ ki o bẹrẹ lẹẹkansi
   • Ayẹwo lẹẹkan ati nigbagbogbo gba abajade
2. Ni awọn ibugbe ti o le rii daju bi iṣiro tabi koodu, kini o le ṣiṣẹ bi ẹsan naa?
   • Awoṣe ere ti o kọ ẹkọ nikan
   • Boya idahun naa tọ tabi awọn idanwo naa kọja
   • Awọn ipari ti awọn esi
   • Awọn nọmba ti àmi ti ipilẹṣẹ
3. Which model family does the cited paper describe using rejection sampling in its alignment pipeline?
   • GPT-4o
   • BERT Base
   • Meta’s Llama 2-Chat
   • Whisper Large
4. Why might a team choose rejection-sampling fine-tuning over a policy-gradient approach for some tasks?
   • It never needs a candidate generator or scoring rule
   • It guarantees no reward hacking
   • It always requires less compute than every RL method
   • It can use supervised learning on selected generations instead of updating through policy-gradient RL
5. Ikẹkọ lori awọn ayẹwo ere ti o pọju ṣe kini?
   • Distills ti o dara ju-ti-N didara sinu awọn abajade nikan-apẹẹrẹ awoṣe
   • Ṣe alekun iwọn fokabulari
   • Yọ awọn awoṣe ká agbara lati se ina ọrọ
   • Ilọpo meji kika paramita
6. What should teams monitor when repeatedly fine-tuning on filtered model-generated outputs?
   • Only the number of model parameters
   • Output diversity alongside held-out correctness
   • Whether every answer uses the same wording
   • Only the training loss on selected samples
7. What is a limitation of rejection-sampling fine-tuning’s candidate selection?
   • It can only be used with image labels
   • It always changes the tokenizer
   • It cannot train on a high-quality answer that was never generated or retained
   • It makes the scoring function unnecessary
8. What risk arises if the scoring model used to filter candidates is flawed?
   • The policy becomes unable to generate any text
   • The tokenizer automatically changes vocabulary
   • The reference labels become unnecessary
   • The policy may learn to exploit the scorer rather than improve the intended quality