Concept迭代式Step-DPO / 过程级偏好优化
Step-DPO
A process-level preference optimization method that constructs preference pairs at the step granularity to provide denser training signals for long reasoning trajectories, thereby improving credit assignment.
Timeline (last 90 days)
Nov 3
MLR introduces iterative Step-DPO, a process-level preference optimization scheme that uses twisted sequential Monte Carlo (TSMC) to provide scalable step-by-step supervision.
Unverified50%