Holonomy_lib, exact non Euclidean geometry primitives for PyTorch
Holonomy_lib is a new PyTorch math library designed for advanced research in differential geometry and related fields. It features a comprehensive set of modules and primitives, all grounded in academic citations. The library aims to streamline the mathematical foundations necessary for modern machine learning applications.
- ▪Holonomy_lib includes twelve modules and 1179 tests, ensuring rigorous validation of its mathematical primitives.
- ▪The library supports various geometric structures, including Riemannian and pseudo-Riemannian manifolds.
- ▪Developed by independent researchers, it emphasizes mechanistic interpretability and content-addressable provenance.
Opening excerpt (first ~120 words) tap to expand
holonomy_lib A research-grade PyTorch math library: GPU-native, batched-first, audit-clean, with every primitive grounded in a citation. Differential geometry, spectral graph theory, discrete Ricci flow, tensor decompositions, Riemannian optimization, simplicial topology, batched persistent homology, and content-addressable provenance for mechanistic interpretability, all under one roof. Developed by independent and Synoros researchers for the substrate research. What this is A consolidated PyTorch math library for research at the intersection of differential geometry, spectral graph theory, computational topology, and mechanistic interpretability: the mathematics that modern ML keeps reinventing project by project.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.