num.cr

Provides an N-dimensional Tensor with support for slicing, mutation, permutation, and broadcasting, enabling complex numerical computations without unnecessary allocations, as shown in the map and reduce examples.

Tensor operations can be backed by OpenCL for significant performance gains on supported hardware, with creation methods allowing CPU or GPU storage initialization.

Leverages established libraries like BLAS, LAPACK, and ClBlast for robust routines such as matrix inversion and eigenvalue computation, demonstrated in the linear algebra code snippets.

Num::Grad offers gradient computation entirely in Crystal, facilitating machine learning workflows without external dependencies, as illustrated in the backpropagation example.