%global debug_package %{nil} Name: llama-cpp-cmpfix Version: 0.6.0 Release: 1%{?dist} Summary: llama.cpp CUDA inference server, CLI, and libraries, with fixes for nvidia cmp cards Conflicts: llama-cpp License: GPL-3.0 URL: https://github.com/Niakr1s/llama.cpp # Build-time only BuildRequires: cmake BuildRequires: gcc-c++ BuildRequires: make BuildRequires: git BuildRequires: openssl-devel BuildRequires: patchelf # Runtime only Requires: cuda-toolkit Source0: %{name}-%{version}.tar.gz %description llama.cpp built with the CUDA backend, with DP4A disabled, that makes inference faster on mining GPUs (like cmp50hx and similar), providing the llama-server HTTP server, the llama-cli command-line tool, and all required shared libraries. %prep %setup -q -n llama.cpp-%{version} %build cmake -B build -DGGML_CUDA=ON \ -DCMAKE_EXE_LINKER_FLAGS="-no-pie" \ -DCMAKE_BUILD_TYPE=Release \ -DCMAKE_CUDA_ARCHITECTURES="75;80;86;89;120" \ -DCMAKE_CUDA_FLAGS="-DDISABLE_DP4A --fmad=false" \ -DCMAKE_INSTALL_RPATH="" cmake --build build -j"$(nproc)" %install mkdir -p %{buildroot}%{_bindir} mkdir -p %{buildroot}%{_unitdir} mkdir -p %{buildroot}%{_libdir} install -m 0755 build/bin/llama-server %{buildroot}%{_bindir}/llama-server install -m 0755 build/bin/llama-cli %{buildroot}%{_bindir}/llama-cli for lib in \ libllama-server-impl.so \ libllama-cli-impl.so \ libllama-common.so \ libmtmd.so \ libllama.so \ libggml.so \ libggml-cpu.so \ libggml-cuda.so \ libggml-base.so ; do cp -a build/bin/${lib}* %{buildroot}%{_libdir}/ done patchelf --remove-rpath %{buildroot}%{_bindir}/llama-server patchelf --remove-rpath %{buildroot}%{_bindir}/llama-cli for lib in %{buildroot}%{_libdir}/libllama*.so* %{buildroot}%{_libdir}/libmtmd*.so* %{buildroot}%{_libdir}/libggml*.so*; do patchelf --remove-rpath "$lib" done %post /usr/sbin/ldconfig for bin in llama-server llama-cli; do /usr/sbin/semanage fcontext -a -t bin_t "%{_bindir}/$bin" 2>/dev/null || : /usr/sbin/semanage fcontext -a -t bin_t "%{_bindir}/$bin" 2>/dev/null || : done %preun %postun /usr/sbin/ldconfig %files %{_bindir}/llama-server %{_bindir}/llama-cli %{_libdir}/libllama-server-impl.so* %{_libdir}/libllama-cli-impl.so* %{_libdir}/libllama-common.so* %{_libdir}/libmtmd.so* %{_libdir}/libllama.so* %{_libdir}/libggml.so* %{_libdir}/libggml-cpu.so* %{_libdir}/libggml-cuda.so* %{_libdir}/libggml-base.so* %changelog %autochangelog