安装llama-cpp-python遇架构不兼容（x86_64而非arm64）求助

问题背景 #

尝试搭建本地高速NLP环境时，无法安装适配arm64架构的llama-cpp-python。执行以下编译安装命令后：

CMAKE_ARGS="-DLLAMA_METAL=on -DLLAMA_METAL_EMBED_LIBRARY=on" \
pip install llama-cpp-python --no-binary llama-cpp-python --force-reinstall --no-cache-dir

导入库时出现架构不兼容错误：

RuntimeError: Failed to load shared library '/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib': dlopen(/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib, 0x0006): tried: '/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64e' or 'arm64')), '/System/Volumes/Preboot/Cryptexes/OS/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib' (no such file), '/Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64e' or 'arm64'))

解决方案 #

1. 确认Python环境架构 #

先检查当前Python的架构类型，避免使用Rosetta转译的x86_64版本：

python3 -c "import platform; print(platform.machine())"

若输出为x86_64，需切换到原生arm64版本的Python（如通过Homebrew安装的arm64 Python）。

2. 强制指定arm64编译架构 #

在安装命令中添加架构指定参数，确保CMake编译出适配arm64的库：

CMAKE_ARGS="-DLLAMA_METAL=on -DLLAMA_METAL_EMBED_LIBRARY=on -DCMAKE_OSX_ARCHITECTURES=arm64" \
ARCHFLAGS="-arch arm64" \
pip install llama-cpp-python --no-binary llama-cpp-python --force-reinstall --no-cache-dir

3. 清理残留编译缓存 #

如果之前有残留的x86_64编译文件，先彻底清理：

rm -rf ~/.cache/pip
rm -rf /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp

清理完成后重新执行步骤2的安装命令。

4. 验证安装结果 #

安装完成后，检查生成的库文件架构是否为arm64：

lipo -info /Library/Frameworks/Python.framework/Versions/3.12/lib/python3.12/site-packages/llama_cpp/lib/libllama.dylib

若输出包含arm64，说明安装成功，此时导入llama_cpp即可正常使用。

内容的提问来源于stack exchange，提问作者Dennis Losett