llama.cpp

Local inference engine

aknowledgements: ggml-org/llama.cpp

Apple Silicon

• set BUILD_SHARED_LIBS to FALSE

Intel

• set GGML_CPU to FALSE
• set CMAKE_OSX_ARCHITECTURES to x86_64

Windows

set LLAMA_CURL to FALSE c.f. ggml-org/llama.cpp#9937

cmake -S . -B build -A x64
 -DLLAMA_STATIC=ON
 -DLLAMA_DIRECTML=ON
 -DCMAKE_TOOLCHAIN_FILE={...\vcpkg\scripts\buildsystems\vcpkg.cmake}
 -DCURL_INCLUDE_DIR={\vcpkg\installed\x64-windows-static\include}
 -DCURL_LIBRARY={\vcpkg\installed\x64-windows-static\lib\libcurl.lib}
 -DLLAMA_BUILD_SERVER=ON
cmake --build build --config Release

• open project sith visual studio
• add curl include paths
• add libraries

Crypt32.lib
Secur32.lib
Iphlpapi.lib
libcurl.lib
zlib.lib
ws2_32.lib

• build each target with MT