mineru-llama-cpp 0.2.0

mineru-llama-cpp 0.2.0 raises the minimum Python version to 3.11 and focuses macOS support on Apple Silicon.

Compatibility changes

• Require Python 3.11 or newer. Wheels now use the cp311-abi3 tag and are tested on standard, GIL-enabled CPython 3.11–3.14.
• Remove macOS Intel support, including x86_64 Python under Rosetta and universal2 builds. These targets are also rejected for source builds.
• Ship five platform wheels: Linux x86_64 and aarch64, Windows AMD64 and ARM64, and macOS arm64. Linux and Windows retain CPU/Vulkan; macOS retains CPU/Metal and the macOS 14 minimum.

The public Python API, inference behavior, and pinned llama.cpp revision are unchanged. The binding includes the explicit C++ object casts required by the Python 3.11 Limited API. Build, publication, and artifact checks enforce the new Python and platform baselines.

Install with:

python -m pip install --upgrade mineru-llama-cpp==0.2.0

Source: 26e30b0. License: MIT.

The migration validation includes a local regression suite with 240 passing tests (including real-model checks) and an identical 19-block page extraction between the old baseline and the CI-built candidate. Full details are in the validation record.

All five platform builds, the older Xcode SDK check, and aggregate platform/stable-ABI audits passed in release CI 39812742859. The project LICENSE and Python/version metadata were checked in all five wheels. An isolated installation of the CI-built macOS arm64 wheel passed import, CPU/Metal backend availability, and short real-model generation checks.

Published to PyPI by workflow 39816955651. All five PyPI file hashes match the attached wheels and SHA256SUMS. The PyPI-hosted macOS arm64 wheel was downloaded, hash-verified, installed into a fresh Python 3.11 environment, and passed import and CPU/Metal backend availability checks.