clifra

Clifra provides layout-first Clifford algebra tools for PyTorch. It represents geometric values as tensors, plans algebra operations from explicit grade layouts, and exposes the resulting work as reusable PyTorch modules. It can be used as a Clifford algebra library, a geometric deep-learning toolkit, or the foundation of a separate research or domain-specific system.

Choose a Path

I want to...	Start here
learn clifra from the beginning	First Clifford Product
understand layouts, planning, or learnable geometry	Explanations
look up a public interface or tensor contract	API Reference
inspect performance and numerical profiles	Benchmarks

Benchmark Suite

The configured benchmark separates full-layout measurements through dimension 8 from compact-layout measurements through dimension 63. It records setup, cold-call, forward and backward timing distributions, throughput, tensor statistics, cumulative error, and the complete execution context.

See Benchmarks for the matrix, commands, graphs, and raw artifacts.

Research Demonstration

The continuum solver brings Clifra's full methodology together: semantic grade layouts, planned Clifford actions, differentiable execution, sampled bivector generators, and objective-driven learning. Its compact example demonstrates the general transformation field, while the physics-informed example develops that field into a validated deformation system.

See Bivector Coordinate Fields for the derivation and the exact sampling, path, and inversion semantics.