AIM enables post-training modulation of large models to change utility levels or focus features without any retraining or additional data.
arXiv · March 16, 2026 · 2603.12755
Why it matters
By redistributing logits based on statistical properties, owners can dynamically control model output quality or feature attention for different users/tiers. This provides a lightweight alternative to maintaining multiple fine-tuned versions of the same backbone model.
From the abstract
Large-scale models are typically adapted to meet the diverse requirements of model owners and users. However, maintaining multiple specialized versions of the model is inefficient. In response, we propose AIM, a novel model modulation paradigm that enables a single model to exhibit diverse behaviors to meet the specific end requirements. AIM enables two key modulation modes: utility and focus modulations. The former provides model owners with dynamic control over output quality to deliver varyin