Nokia Open-Sources AnyJev, Library to Transform Any LLM into Fast Decision Model

Nokia Applied Research has released AnyJev, an open-source Python library that converts any open-weight language model into a calibrated decision system without requiring model training or retraining. The library, licensed under Apache 2.0, extracts structured decisions—multiple-choice selections, yes-or-no judgments, and probability scores—directly from a model's internal token predictions during a single pass, then applies zero-label cyclic-shift marginalization to eliminate option-order bias and improve confidence calibration.