File ollama.spec of Package ollama
#
# spec file for package ollama
#
# Copyright (c) 2026 SUSE LLC and contributors
#
# All modifications and additions to the file contributed by third parties
# remain the property of their copyright owners, unless otherwise agreed
# upon. The license for this file, and modifications and additions to the
# file, is the same license as for the pristine package itself (unless the
# license for the pristine package is not an Open Source License, in which
# case the license is the MIT License). An "Open Source License" is a
# license that conforms to the Open Source Definition (Version 1.9)
# published by the Open Source Initiative.
# Please submit bugfixes or comments via https://bugs.opensuse.org/
#
%global rocm_major 7
%global rocm_minor 2
%global rocm_release %{rocm_major}.%{rocm_minor}
%if 0%{?is_opensuse}
%{bcond_with cuda}
%else
%{bcond_without cuda}
%endif
%{bcond_without rocm}
%if 0%{?sle_version} && 0%{?sle_version} >= 150600
%global force_gcc_version 12
%endif
%if 0%{?suse_version} >= 1699
%global force_gcc_version 14
%endif
%define cuda_version_major 13
%define cuda_version_minor 0
%define cuda_version %{cuda_version_major}-%{cuda_version_minor}
Name: ollama
Version: 0.17.6
Release: 0
Summary: Tool for running AI models on-premise
License: MIT
URL: https://ollama.com
Source: https://github.com/ollama/ollama/archive/v%{version}/%{name}-%{version}.tar.gz
Source1: vendor.tar.zstd
Source2: %{name}.service
Source3: %{name}-user.conf
Source4: sysconfig.%{name}
Patch0: fix-mlxrunner-tests.diff
Patch1: ollama-support-gfx1103.patch
Patch2: ollama-no-hip-redist.patch
Patch3: ollama-no-vulkan-redist.patch
Patch4: ollama-lib64-runner-path.patch
BuildRequires: cmake >= 3.24
BuildRequires: git-core
BuildRequires: ninja
BuildRequires: pkgconfig
BuildRequires: shaderc
BuildRequires: sysuser-tools
BuildRequires: zstd
BuildRequires: golang(API) >= 1.24
BuildRequires: group(render)
BuildRequires: group(video)
BuildRequires: pkgconfig(vulkan)
Requires: group(render)
Requires: group(video)
Recommends: ( %{name}-vulkan or %{name}-cuda or %{name}-rocm )
%if %{with cuda}
# requires cuda-toolkit*-config-common, cuda-cudart, cuda-cccl
BuildRequires: cuda-cudart-devel-%{cuda_version}
BuildRequires: cuda-driver-devel-%{cuda_version}
# requires cuda-crt, cuda-nvvm
BuildRequires: cuda-nvcc-%{cuda_version}
# requires libcublas
BuildRequires: libcublas-devel-%{cuda_version}
%endif
# %if %{with rocm}
# BuildRequires: hipblas-devel
# BuildRequires: rocblas-devel
# BuildRequires: rocm-comgr-devel
# BuildRequires: rocm-compilersupport-macros
# BuildRequires: rocm-hip-devel
# BuildRequires: rocm-rpm-macros
# BuildRequires: rocm-runtime-devel
# BuildRequires: rocminfo
# %endif
Requires(pre): %fillup_prereq
# 32bit seems not to be supported anymore while riscv isn't yet.
ExcludeArch: %{ix86} %{arm} riscv64
%sysusers_requires
%if 0%{?force_gcc_version}
BuildRequires: gcc%{force_gcc_version}-c++
BuildRequires: libstdc++6-devel-gcc%{?force_gcc_version}
%else
%if 0%{suse_version} >= 1600
BuildRequires: gcc-c++ >= 11.4.0
%else
BuildRequires: gcc14-c++
%endif
%endif
Requires(pre): %{fillup_prereq}
%if %{with rocm}
BuildRequires: rocm-release(hip-devel) = %{rocm_release}
BuildRequires: rocm-release(rpm-macros) = %{rocm_release}
BuildRequires: rocm-release(clang) = %{rocm_release}
BuildRequires: rocm-release(lld) = %{rocm_release}
BuildRequires: rocm-release(llvm) = %{rocm_release}
BuildRequires: rocm-release(rocblas-devel) = %{rocm_release}
BuildRequires: rocm-release(rocm-runtime-devel) = %{rocm_release}
BuildRequires: rocm-release(rocprim-headers) = %{rocm_release}
BuildRequires: rocm-release(rocsolver-devel) = %{rocm_release}
BuildRequires: rocm-release(rocsparse-devel) = %{rocm_release}
BuildRequires: cmake(COMgr-ROCm) = %{rocm_release}
BuildRequires: rocm-release(hipblas-common-headers) = %{rocm_release}
BuildRequires: rocm-release(hipblas-devel) = %{rocm_release}
BuildRequires: pkgconfig(libdrm)
BuildRequires: pkgconfig(libdrm_amdgpu)
%endif
%description
Ollama is a tool for running AI models on one's own hardware.
It offers a command-line interface and a RESTful API.
New models can be created or existing ones modified in the
Ollama library using the Modelfile syntax.
Source model weights found on Hugging Face and similar sites
can be imported.
%package vulkan
Summary: Ollama Module using Vulkan
Group: System/Libraries/C and C++
Requires: %{name}%{?_isa} = %{version}-%{release}
%description vulkan
Ollama plugin module using Vulkan.
%package cuda
Summary: Ollama Module using CUDA
Group: System/Libraries/C and C++
Requires: %{name}%{?_isa} = %{version}-%{release}
%description cuda
Ollama plugin module using NVIDIA CUDA.
%package rocm-%{rocm_major}-%{rocm_minor}
Summary: Ollama Module using AMD ROCm
Group: System/Libraries/C and C++
Requires: %{name}%{?_isa} = %{version}-%{release}
Requires: rocm-release(clang-libs) = %{rocm_release}
Requires: rocm-release(llvm-libs) = %{rocm_release}
Requires: rocm-release(rocblas) = %{rocm_release}
Requires: rocm-release(rocm-runtime) = %{rocm_release}
Requires: rocm-release(rocsolver) = %{rocm_release}
Requires: rocm-release(rocsparse) = %{rocm_release}
Requires: rocm-release(hipblas) = %{rocm_release}
Provides: rocm-release(ollama-runner) = %{rocm_release}
Conflicts: rocm-release(ollama-runner) < %{rocm_release}
%description rocm-%{rocm_major}-%{rocm_minor}
This package contains the shared library for running
LLMs on AMD GPUs using ROCm version %{rocm_release}.
%prep
%autosetup -a1 -p1 -n %{name}-%{version}
%build
%define __builder ninja
%sysusers_generate_pre %{SOURCE3} %{name} %{name}-user.conf
export GOFLAGS="-buildmode=pie -mod=vendor -v"
export CXX="g++%{?force_gcc_version:-%force_gcc_version}"
export CC="gcc%{?force_gcc_version:-%force_gcc_version}"
export GCCDIST=`$CXX -dumpmachine`
#export GOFLAGS="-mod=vendor -v"
%if %{with cuda}
export PATH=/usr/local/cuda-%{cuda_version_major}.%{cuda_version_minor}/bin:$PATH
export LIBRARY_PATH=/usr/local/cuda/lib64/stubs
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/cuda/lib64
export LDFLAGS=-s
export GOFLAGS="${GOFLAGS} '-ldflags=-w -s -X=github.com/ollama/ollama/version.Version=%{version} -X=github.com/ollama/ollama/server.mode=release\'"
export CGO_ENABLED=1
# NOTE: NVIDIA compiler (nvcc) will look for the default g++ so we
# have to set it to g++12 explicitly
export CUDAHOSTCXX=/usr/bin/g++%{?force_gcc_version:-%{force_gcc_version}}
%endif
# # fix CMAKE_BINARY_DIR / OLLAMA_INSTALL_DIR location
# sed -i -e 's@\(${CMAKE_BINARY_DIR}\)/lib/ollama@\1/%{_lib}/ollama@' CMakeLists.txt
# sed -i -e 's@\(${CMAKE_INSTALL_PREFIX}\)/lib/ollama@\1/%{_lib}/ollama@' CMakeLists.txt
# # overwrite ml/path.go so LibOllamaPath is set to our path
# echo -e 'package ml\nvar LibOllamaPath string = "/usr/%{_lib}/ollama"' > ml/path.go
# # for dlopens
# sed -i -e 's@"lib"@"%{_lib}"@' \
# -e 's@"lib/ollama"@"%{_lib}/ollama"@' \
# ml/backend/ggml/ggml/src/ggml.go
%cmake \
-UCMAKE_INSTALL_BINDIR -DCMAKE_INSTALL_BINDIR=%{_libdir}/ollama \
-UOLLAMA_INSTALL_DIR -DOLLAMA_INSTALL_DIR=%{_libdir}/ollama \
-DGGML_BACKEND_DIR=%{_libdir}/ollama \
%{?with_cuda:-DCMAKE_CUDA_COMPILER=/usr/local/cuda-%{cuda_version_major}.%{cuda_version_minor}/bin/nvcc} \
%{?with_rocm:-DCMAKE_HIP_COMPILER=clang-rocm -DCMAKE_HIP_FLAGS="%{?force_gcc_version:--gcc-install-dir=%{_libdir}/gcc/$GCCDIST/%{force_gcc_version}}" \
-DCMAKE_HIP_COMPILER_ROCM_LIB=%{_libdir} -DAMDGPU_TARGETS="%{list_sep rocm_gpu_list ;}"} \
%{nil}
%cmake_build
cd ..
go build -trimpath -o %{name} .
%install
export GOFLAGS="-buildmode=pie -mod=vendor -v"
export CXX="g++%{?force_gcc_version:-%force_gcc_version}"
export CC="gcc%{?force_gcc_version:-%force_gcc_version}"
%cmake_install
install -D -m 0755 %{name} %{buildroot}/%{_bindir}/%{name}
install -D -m 0644 %{SOURCE2} %{buildroot}%{_unitdir}/%{name}.service
install -D -m 0644 %{SOURCE3} %{buildroot}%{_sysusersdir}/%{name}-user.conf
install -D -m 0644 %{SOURCE4} %{buildroot}%{_fillupdir}/sysconfig.%{name}
install -d %{buildroot}%{_localstatedir}/lib/%{name}
mkdir -p %{buildroot}%{_sbindir}
ln -s /usr/sbin/service %{buildroot}%{_sbindir}/rc%{name}
mkdir -p "%{buildroot}/%{_docdir}/%{name}"
cp -Ra docs/* "%{buildroot}/%{_docdir}/%{name}"
# remove copies of system libraries
shopt -s nullglob
rm -rf %{buildroot}%{_libdir}/%{name}/lib{vulkan,drm,amd,hip,cu,hsa,roc,numa,elf}*
%check
export GOFLAGS="-buildmode=pie -mod=vendor -v"
export CXX="g++%{?force_gcc_version:-%force_gcc_version}"
export CC="gcc%{?force_gcc_version:-%force_gcc_version}"
go test -v ./...
%pre -f %{name}.pre
%service_add_pre %{name}.service
%post
%service_add_post %{name}.service
%fillup_only
%preun
%service_del_preun %{name}.service
%postun
%service_del_postun %{name}.service
%files
%doc README.md
%license LICENSE
%dir %{_libdir}/%{name}/
%{_docdir}/%{name}
%{_bindir}/%{name}
%{_libdir}/%{name}
%{_sbindir}/rc%{name}
%{_unitdir}/%{name}.service
%{_sysusersdir}/%{name}-user.conf
%exclude %{_libdir}/ollama/libggml-cuda.so
%exclude %{_libdir}/ollama/libggml-hip.so
%exclude %{_libdir}/ollama/libggml-vulkan.so
%{_fillupdir}/sysconfig.%{name}
%attr(-, ollama, ollama) %{_localstatedir}/lib/%{name}
%files vulkan
%{_libdir}/ollama/libggml-vulkan.so
%if %{with cuda}
%files cuda
%{_libdir}/ollama/libggml-cuda.so
%endif
%if %{with rocm}
%files rocm-%{rocm_major}-%{rocm_minor}
%{_libdir}/ollama/libggml-hip.so
%endif
%changelog
