All Tools
L
Dev ToolsFreeOpen Source
LINEAX
Differentiable linear solvers for JAX and Equinox
Apache-2.0
ABOUT
JAX models often need least-squares or iterative linear solves that stay on device and through autodiff. SciPy linalg leaves the accelerator. Lineax wraps those solvers so Equinox training loops invert systems without a host copy.
INSTALL
pip install lineaxINTEGRATION GUIDE
1. Solve a batched least-squares problem inside a JAX loss
2. Invert a linear system with a GPU-resident iterative solver
3. Differentiate through a Lineax solve in an Equinox model
4. Replace a NumPy linalg call that was copying off device
TAGS
pythonjaxlinear-algebraequinoxleast-squaresscientific