llama.cpp

Local inference engine (repository name changed to comply with dependency manager)

[!WARNING] Windows CLI is built without libcurl. That means network feature such as

llama-server -hf elyza/Llama-3-ELYZA-JP-8B-GGUF:Q4_K_M

only works on Mac.

Note for CMake

for x86_64

• set GGML_CPU to FALSE
• set CMAKE_OSX_ARCHITECTURES to x86_64

for Windows

• set LLAMA_CURL to FALSE

c.f. https://github.com/ggml-org/llama.cpp/issues/9937

for All

• set BUILD_SHARED_LIBS to FALSE