Changelog

0.4.0

• Upgrade to jax>=0.5.0
• Fix issues in vmap
• Fix issues in jacfwd/jacrev
• No more hidden segfaults (hopefully)
• More consistent array shape broadcasting
• Drop support for x86 MacOS (not supported by jax>=0.5.0 either)
• Update GitHub CI

0.3.1

• Bugfixes

0.3.0

• Implement new FFI API for C++ extension
• Enhance C++ testing and error handling (proper error throwing instead of segfaulting)
• Enable JIT and vmap for optimized performance
• Improve shape handling for arrays
• Retrieve array sizes from C++ buffers
• Refactor workflows and remove deprecated implementations
• Remove old notebook files
• Upgrade dependencies and ensure compatibility with C++17 standard

0.2.10

• Run tests post wheel build
• Pin exact dependency versions
• Streamline GitHub workflows and update setup.py
• Fix issues with vmap over array b

0.2.8

• Refine CI/CD environment variable configuration for cibuildwheel

0.2.7

• Prevent memory leaks
• Introduce pre-commit configuration
• Update cibuildwheel configuration
• Clone specific SuiteSparse version

0.2.5

• Address deprecations in XLA translations

0.2.4

• Add support for Python 3.12
• Consolidate GitHub workflow files and update package metadata

0.2.0

• Vendor SuiteSparse library in source distribution
• Re-enable PIP_FIND_LINKS
• Update build recipes and dependencies
• Improve README with setup and build instructions

0.1.4

• Add support for Python 3.11

0.1.3

• Enable installation on macOS
• Fix issues with static linking on macOS (C++11 requirement)

0.1.1

• Publish release on PyPI and include tarball for distribution
• Add support for multiple Python versions and manylinux2014 wheels

0.1.0

• Enable custom JVP/VJP rules
• Improve differentiation features with forward-mode JVP and transposition
• Add pyproject.toml for better build configuration

0.0.6

• Add more library/include paths for builds
• Refine README and setup instructions
• Initial integration of complex value handling in vmap

0.0.4

• Add bump2version configuration for automated versioning
• Bugfix: Correct matrix-vector multiplication in COO format (mul_coo_vec)

0.0.3

• Set up core functionality with sparse matrix multiplication (COO format)
• Integrate vmap for float64 and complex128 arrays
• Initial setup with Makefile, test suites, and Docker configuration
• Begin development of XLA translations and gradient support