Show HN: Wazn-2B – decision model with separate encoding and competition (huggingface.co)

Wazn-2B is a novel decision model designed to process contextual prompts and return probability distributions over candidate labels, utilizing a unique framework that incorporates separate encoding and relational comparison. The model, powered by the Qwen3.5-2B backbone with LoRA adapters, encodes each candidate label along with relevant rules and examples, and employs a relational judge to compare these candidates rather than score them in isolation. Notably, it also includes a "NONE" gate to evaluate whether any of the candidates are applicable, enhancing its decision-making capabilities. This project is significant for the AI and ML community as it introduces a new approach to decision modeling that could improve the accuracy and efficiency of label selection in complex contexts. While currently in an experimental phase and not intended for production use, the model offers a promising framework for future development. The emphasis on relational comparisons and candidate validity checks presents a shift from traditional scoring to a more nuanced evaluation of options, making Wazn-2B a potential game-changer in decision-making applications within AI systems.