Riemannian-Lorentz Fusion for Heterogeneous Model Merging
September 18, 2026
The Riemannian-Lorentz Parameter Fusion (RLPF) method enables merging Vision Transformers (ViT) with State-Space Models (SSM). It uses hyperbolic space geometry to align different architectures by semantic role, bypassing the need for shape-compatible checkpoints.
HOW THIS AFFECTS YOU
●
builderYou can leverage this to combine the strengths of different model types without expensive retraining.
●
researcherThis offers a mathematical framework for merging fundamentally different architectural paradigms.