geoopt/geoopt

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1091 stars · 94 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Geoopt is a PyTorch library providing manifold-aware optimizers and manifold tensor containers for Riemannian optimization. It implements Riemannian Adaptive Optimization Methods (ICLR 2019) plus a variety of manifolds such as the Poincare ball, sphere, Stiefel, and hyperboloid models.

Use cases

• optimize neural network parameters constrained to a Riemannian manifold in PyTorch
• train models with hyperbolic embeddings using the Poincare ball
• perform gradient descent on the sphere or Stiefel manifold
• implement Riemannian Adam and SGD optimizers
• project tensors and gradients onto manifolds in deep learning pipelines

When to choose

• you need Riemannian optimizers integrated with torch.optim
• your model parameters live on manifolds like spheres, hyperbolic spaces, or Stiefel matrices
• you want manifold-aware tensors that work with standard PyTorch modules

When to avoid

• you only need standard Euclidean optimization
• you need a fully stable, frozen API - the project is under active development and the API may change
• you use a PyTorch version older than 2.0.1

Facets

library · maturity active

machine-learning math machine-learning deep-learning mathematics python riemannian-optimization pytorch manifolds hyperbolic-embeddings optimizers optimization algorithms

Member repositories

For agents

markdown · JSON · MCP: product_card(name="geoopt/geoopt")