%FILENAME%
llama.cpp-cuda-git-b9967.r9.e3546c7948-1-x86_64.pkg.tar.zst

%NAME%
llama.cpp-cuda-git

%BASE%
llama.cpp-cuda-git

%VERSION%
b9967.r9.e3546c7948-1

%DESC%
Port of Facebook's LLaMA model in C/C++ (with NVIDIA CUDA optimizations)

%CSIZE%
8955822

%ISIZE%
19708130

%SHA256SUM%
10bc65c9ad60a03ba8ce411c464dbe370a775306d8ca127c4595541d1fc0e3ab

%URL%
https://github.com/ggml-org/llama.cpp

%LICENSE%
MIT

%ARCH%
x86_64

%BUILDDATE%
1783812066

%PACKAGER%
Unknown Packager

%CONFLICTS%
llama.cpp

%PROVIDES%
llama.cpp

%DEPENDS%
ggml-cuda-git
curl
gcc-libs
glibc
openssl

%OPTDEPENDS%
ccache: greatly reduce package re-build time
nccl: needed for multi-GPU parallelism
python-numpy: needed for convert_hf_to_gguf.py
python-safetensors: needed for convert_hf_to_gguf.py
python-sentencepiece: needed for convert_hf_to_gguf.py
python-pytorch: needed for convert_hf_to_gguf.py
python-transformers: needed for convert_hf_to_gguf.py
rdma-core: RDMA transport for RPC backend

%MAKEDEPENDS%
cmake
cuda
git
ninja

