v1.0.0
CVXPYlayers 1.0
Major rewrite of CVXPYlayers with a new architecture and expanded framework support.
Highlights
- New unified architecture — Refactored core with
LayersContextdataclass and solver interface abstractions - MLX support — New Apple MLX backend alongside PyTorch and JAX
- Multiple solver backends — Support for diffcp (default), CuClarabel (GPU), MPAX, and Moreau
- Dual variable support —
CvxpyLayercan now return dual variables from constraints - Geometric program (GP) support — Full DGP-to-DCP reduction with batched execution
torch.compileandjax.jit+vmapsupport for the Moreau interface- Comprehensive test suite with ruff linting and pre-commit hooks
- Documentation site with Sphinx, example notebooks, and autodoc
Breaking Changes
- Requires Python >= 3.11
- Requires CVXPY >= 1.7.4
- New source layout under
src/cvxpylayers/