%FILENAME%
llama.cpp-cuda-git-b10488.r10.60addddf3c-1-x86_64.pkg.tar.zst

%NAME%
llama.cpp-cuda-git

%BASE%
llama.cpp-cuda-git

%VERSION%
b10488.r10.60addddf3c-1

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

%CSIZE%
9548851

%ISIZE%
20570073

%SHA256SUM%
f2899030cf6bae55ecc2d51f8d3c613c6a2f9b918cceaa9800b0ed33d5175865

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

%LICENSE%
MIT

%ARCH%
x86_64

%BUILDDATE%
1787093602

%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

%MAKEDEPENDS%
cmake
cuda
git
ninja

