xolky (/ˈʃɔl.ki/)

Solve linear systems using sparse Cholesky decomposition with cuDSS in Jax within jitted functions.

Installation

First, install CuDSS (https://developer.nvidia.com/cudss-downloads) and then you can install xolky.

pip install git+ssh://git@github.com/pedrozudo/xolky.git

Examples

Go to the examples directory 🙂

What is Supported?

Operating System

• ✅ Linux
• ❌ Windows
• ❌ macOS

Which Higher Order Functions?

• ✅ jit
• ❌ grad
• ❌ vmap
• ❌ pmap

What Part of the cuDSS API?

• ✅ solving sparse positive definite linear systems
• ❌ all the rest

Precision?

• ✅ fp32 (uses fp64 on the cuDSS side)
• ✅ fp64