Metadata-Version: 2.4
Name: fastmace-oeq-ext
Version: 0.7.0.post1
Summary: Prebuilt OpenEquivariance 0.7.0 host extension for FastMACE (torch 2.14.1+cu130), no C++ compiler needed
Author: Fabian Zwingel
License-Expression: MIT AND BSD-3-Clause
Project-URL: Source, https://github.com/FZWINGEL-UBT/FastMACE
Project-URL: OpenEquivariance, https://github.com/PASSIONLab/OpenEquivariance
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Operating System :: POSIX :: Linux
Classifier: Environment :: GPU :: NVIDIA CUDA :: 13
Requires-Python: <3.14,>=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
License-File: fastmace_oeq_ext/LICENSE.openequivariance
Requires-Dist: torch==2.14.1
Requires-Dist: openequivariance==0.7.0
Dynamic: license-file

# fastmace-oeq-ext

OpenEquivariance 0.7.0's host extension (`libtorch_tp_jit.so`), prebuilt for torch 2.14.1+cu130 on
Linux x86_64 (manylinux_2_28, glibc 2.28 or newer), CPython 3.10 to 3.13. Installed by
`pip install "fastmace[oeq]"`, it lets FastMACE use the hybrid cuEquivariance/OpenEquivariance backend
without g++, ninja or nvcc. FastMACE checks `manifest.json` and falls back to building the extension
itself when torch or OpenEquivariance differ from the versions it was built for.
A C compiler is still needed by Triton's launcher stub; GPU kernels continue to compile at run time.

The library is built unmodified from OpenEquivariance's sources (BSD 3-Clause, The Regents of the
University of California, LBNL; see `LICENSE.openequivariance`). The packaging is MIT (FastMACE).
