Video
Unknown · 0:00
ORPO (Monolithic Preference Optimization without Reference Model, from KAIST AI) does supervised fine-tuning and preference alignment in one training step instead of two. It adds an odds-ratio loss to the standard SFT...
Unknown · 0:00
ORPO (Monolithic Preference Optimization without Reference Model, from KAIST AI) does supervised fine-tuning and preference alignment in one training step instead of two. It adds an odds-ratio loss to the standard SFT...
Redirecting...