Tag
GeoSteer is an optimization-based method for norm-preserving activation steering in large language models, using geodesic updates on the representation manifold to improve control over model behavior while preserving activation norms.
Norm-preserving abliteration technique applied to Qwen3.6-35B-A3B achieves 0% refusal rate while maintaining benchmark performance, with open source dataset released.
The paper introduces YB-Mixer, a token-mixing layer derived from the generalized Yang-Baxter equation, which is exactly norm-preserving, depth-stable, and allows order-free and variable-budget inference. It achieves competitive performance on long-range memory tasks with fewer parameters compared to attention and state-space baselines.