[KongchangAI]
Conceptpost-training shift

Post-training Shift

The difference in log-probability between a teacher model and a reference policy, reflecting the behavioral preferences the teacher acquired through post-training such as RLHF or rejection-sampling fine-tuning.

Timeline (last 90 days)

Nov 6

D²-OPD determines the extrapolation direction by adaptively weighted combination of the teacher model's post-training offset and the initial student-teacher gap.

Unverified50%

All Facts (1)

Source Articles