Nokia unveils open-source tool to transform llms into decision models without training

Nokia has released AnyJev, an open-source Python library that converts any open large language model (LLM) into a calibrated decision model without requiring training. The tool is designed for production environments where the goal is to select one answer from a fixed set rather than generate a full sentence. AnyJev leverages the interface of Jev, a decision model launched by TypeSafe AI in September 2026. It provides a probability score for each option, allowing users to set a threshold for decision-making. The library supports three question types and addresses two flaws in existing methods: sensitivity to option order and uncalibrated probabilities. The tool uses levels, with L0 applying fixes through prefills and L1 adding temperature scaling. Testing on Qwen3-8B with BANKING77 showed promising results. The library is available on PyPI, under Apache-2.0, and supports transformers and vLLM backends. Nokia encourages users to check the GitHub repository and engage with their community on social platforms. The project highlights Nokia’s commitment to advancing decision-making capabilities in LLMs through open-source innovation. By eliminating the need for training, AnyJev offers a scalable solution for deploying reliable decision models in real-world applications. The tool’s performance metrics and technical specifications underscore its potential to enhance the efficiency and accuracy of automated decision systems.