%FILENAME%
beellama.cpp-cuda-0.3.1-1-x86_64.pkg.tar.zst

%NAME%
beellama.cpp-cuda

%BASE%
beellama.cpp-cuda

%VERSION%
0.3.1-1

%DESC%
DFlash & TurboQuant fork of llama.cpp — up to 3x faster generation and 7.5x more KV cache (CUDA)

%CSIZE%
56732386

%ISIZE%
95652823

%SHA256SUM%
8f10c3b81235d15b0d509c0dbf89fbd016795652ebdb50d8f62848d844aac49a

%URL%
https://github.com/Anbeeld/beellama.cpp

%LICENSE%
MIT

%ARCH%
x86_64

%BUILDDATE%
1782968292

%PACKAGER%
Unknown Packager

%CONFLICTS%
beellama.cpp
beellama.cpp-bin
llama.cpp
libggml
ggml

%PROVIDES%
beellama.cpp

%DEPENDS%
cuda
curl
gcc-libs
glibc
nvidia-utils
openssl

%OPTDEPENDS%
python-numpy: convert_hf_to_gguf.py
python-safetensors: convert_hf_to_gguf.py
python-sentencepiece: convert_hf_to_gguf.py
python-pytorch: convert_hf_to_gguf.py
python-transformers: convert_hf_to_gguf.py
rdma-core: RDMA transport for RPC backend

%MAKEDEPENDS%
cmake
gcc14
ninja

