llama-cpp dockerized build

A docker-based setup to build llama-cpp binaries:

• supports CUDA and CPU based builds
• cross-build for Linux x86_64 and arm64

Used as a base for nixiesearch/nixiesearch embedded llamacpp-server.

Usage

Create the build container:

docker build -t llamacpp-build:latest --platform=x86_64 .

Build the binary without CUDA:

docker run -i -t -e "LLAMACPP_GPU=false" -v ./llama.cpp:/llama.cpp llamacpp-server-bin:latest

OR with CUDA:

docker run -i -t -e "LLAMACPP_GPU=true" -v ./llama.cpp:/llama.cpp llamacpp-server-bin:latest

Binaries

Resulting binaries are going to be found in llama.cpp/build/bin

License

Apache 2.0