quiz bu lëkkale ak tegtal · Diggante Niveau
Quiz modeling neexal
Test neexal-model poñ, done taamu, pairwise rang perte, neexal piratage, ak gëna-ci-N jëfandikoo ci RLHF. Saytu xalaat yi ak yaatuwaayam.
Laaj 1 de 8
Lan mooy njariñu njëkk ci xeetu neexal ngir ñaari tontu yu gaaw yuñ jox?
Xoolaatal laaj bu nekk8 laaj
Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.
- Lan mooy njariñu njëkk ci xeetu neexal ngir ñaari tontu yu gaaw yuñ jox?
- Benn poñ scalar
- Tontub bind bu mat sëkk
- Séddale probabilite ci vokabileer bi
- Limu token yiñ taamu
- What feedback format did the InstructGPT paper collect to train its reward model?
- A guaranteed correctness label for every possible answer
- Human rankings of several model responses, converted into pairwise preferences
- A single sentiment score for each training prompt
- Only binary safety labels with no response comparisons
- Lan mooy gëna xéewale xeetu neexal buñ miin?
- Valeur absolu bu bépp tontu
- Token bi ci topp
- Ni tontu biñ tànn moo ëpp poñ yi ñu bañ
- Guddaayig tontu yi
- Luy 'pirateri neexal'?
- Sàcc xeetu neexal bu benn concurrent
- ab pexe ngir gaawlu taggat yaram
- Encrypter siñaalu neexal bi
- Politigu jëfandikoo fi yaxxu ci xeetu neexal ngir am poñ yu bari te amul kalite bu dëggu
- Ci wàllu architecture, naka lañuy jëlee xeetu neexal ci xeetu làkk?
- Ni ñuy wecci boppu token-prediction ak benn boppu génne scalar
- Soo yokke yeneen xeeti bàyyi xel
- Soo dindi bépp couche transformateur
- Soo yokke vocabulaire bi ñaari yoon
- naka lañuy jëfandikoo xeetu neexal ci waxtu inference ci sampling bi gëna baax ci N?
- Taggat N xeetu neexal ci paralel
- Defar tontu yu bari ak jëfandikoo xeetu neexal ngir tànn bi gëna am poñ
- Tann N token yiñ gëna xam
- Moyenne N model de contrôle
- Why might an RLHF study use an ensemble of reward models during training?
- To guarantee that reward hacking is impossible
- To replace the human preference data with exact correctness labels
- To estimate disagreement or uncertainty across learned preference scorers
- To ensure every candidate receives the same score
- Why should reward-model evaluations include held-out comparisons?
- To ensure training comparisons are never used
- To make every output identical
- To guarantee the reward is unbiased
- To test whether preferences generalize beyond the training comparisons