DiffSharp

DiffSharp is an open-source tensor library aimed at enabling advanced differentiable programming, primarily for machine learning, probabilistic programming, and optimization tasks. It provides powerful and composable automatic differentiation, supporting nested and mixed-mode differentiation up to any level. Built with F# and targeting .NET, DiffSharp offers APIs and features similar to PyTorch, plus a robust backend leveraging LibTorch and CUDA for accelerated computing on Linux, macOS, and Windows. It is ideal for developers, researchers, and academics looking for high-performance, functional tensor computation and differentiation on the .NET stack.