Cross-platform F32 tensor library — matrix foundation for culebra. Lazy graph + fusion, Accelerate/Metal backends (own CPU/CUDA planned). - View it on GitHub