Welcome to the Documentation of Flint

Why Flint?

Because i like to develop important software with easy deployable open source and fast libraries.
Developing in Python may be fine for small research projects, but it is nothing that should be used in performance critical applications.

Training a network in Python and then rewriting everything for a faster language wastes programmer salary.
Why not just use a machine learning framework that feels as easy as programming Python code in a language like C++?

Example

Tensor<int, 3> t1{{{0}, {1}}, {{2}, {3}}};
Tensor<int, 2> t2 = t1.repeat(1, 2).flattened(2);
std::cout << t2 << std::endl;
/* Tensor<INT32, shape: [4, 6]>( 
 *  [[0, 1, 0, 1, 0, 1], 
 *   [2, 3, 2, 3, 2, 3], 
 *   [0, 1, 0, 1, 0, 1], 
 *   [2, 3, 2, 3, 2, 3]]) */

Efficiency

Flint comes with two backends.
The first one is a CPU backend with a thread pool.
But the real performance is unlocked with the GPU Backend.
Don't feel the necessity to buy CUDA capable devices because we
deploy OpenCL to boost the computation on nearly any available accelerator.