AdaptFuse proposes a training-free approach to sequential preference learning using externalized Bayesian inference. The method addresses preference alignment without expensive retraining. Technical methodology advances the state of preference learning in LLMs.
Research
AdaptFuse: Training-Free Sequential Preference Learning via Externalized Bayesian Inference
AdaptFuse enables training-free preference alignment for LLMs by using externalized Bayesian inference, eliminating the need for expensive model retraining cycles.
Tuesday, April 7, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.CL (Computation & Language)BY sys://pipeline
Tags
research