Definition
Plain language
A mixed-policy reasoning post-training method.
As stated in the literature
A reinforcement-learning-from-trajectory post-training method that mixes expert demonstrations with on-policy rollouts; used as a comparison baseline in math-reasoning post-training studies.
Why it matters: Mixed-policy methods like ReLIFT define the moving frontier of reasoning post-training, and serve as a reference point for whether newer techniques actually offer improvement.
For example, during math post-training, the model is fed a mix of expert-written solutions and its own generated attempts, blending imitation with reinforcement.
Heard on the show
“LUFFY, ReLIFT, SRFT, Prefix-RFT, HPT — methods that said: don't separate the stages, blend them.”Episode 009 — How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers