Metodebevisregistrering

Direct Preference Optimization

Direct Preference Optimization (DPO) is a training method introduced by Rafailov et al. in 2023 that aligns language models with human preferences without requiring an explicit reward model. By directly optimizing for preference pairs (better response vs worse response), DPO simplifies the training pipeline compared to reinforcement learning from human feedback (RLHF).

Sources recorded, not reviewed

Kilderegistrering

Citater kopieret ordret fra metodens kilderegistrering. Ingen påstandsniveauverifikation er udledt heraf.

Direct Preference Optimization: Your Language Model is Secretly a Reward Model

Taksonomisk metoderegistrering · ml-model / deep-learning

Rafailov, R., Sharma, A., Mitchell, E., Manning, C. D., Ermon, S., & Finn, C. (2023). Direct preference optimization: Your language model is secretly a reward model. arXiv preprint arXiv:2305.18290. · URL

Åbn fuld metode

Kuraterede påstande

Påstande gemt i bevis-loggen, hver med sin egen vurdering.

Ingen kuraterede påstande endnu

Denne visning opfinder ikke en påstandsvurdering, når loggen ingen har.

Relaterede metoder

Genereret fra metodegrafen og vist som maskinelt foreslåede relationer — ingen bevispåstand er udledt.

Same method familyLatent Diffusion Modelsmachine-suggested · Relational suggestion, not evidence.Same method familyMamba (State Space Model)machine-suggested · Relational suggestion, not evidence.Same method familyMasked Autoencodersmachine-suggested · Relational suggestion, not evidence.Same method familyQLoRAmachine-suggested · Relational suggestion, not evidence.