From: JohnnyB Date: Mon, 28 Aug 2023 06:31:24 +0000 (+0100) Subject: devops : added systemd units and set versioning to use date. (#2835) X-Git-Tag: gguf-v0.4.0~209 X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=3e8ff47af620a31e0810c58a41e4b089145982ef;p=pkg%2Fggml%2Fsources%2Fllama.cpp devops : added systemd units and set versioning to use date. (#2835) * Corrections and systemd units * Missing dependency clblast --- diff --git a/.devops/lamma-cpp-clblast.srpm.spec b/.devops/lamma-cpp-clblast.srpm.spec deleted file mode 100644 index 739c6828..00000000 --- a/.devops/lamma-cpp-clblast.srpm.spec +++ /dev/null @@ -1,58 +0,0 @@ -# SRPM for building from source and packaging an RPM for RPM-based distros. -# https://fedoraproject.org/wiki/How_to_create_an_RPM_package -# Built and maintained by John Boero - boeroboy@gmail.com -# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal - -# Notes for llama.cpp: -# 1. Tags are currently based on hash - which will not sort asciibetically. -# We need to declare standard versioning if people want to sort latest releases. -# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies. -# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed. -# Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo -# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries. -# It is up to the user to install the correct vendor-specific support. - -Name: llama.cpp-clblast -Version: master -Release: 1%{?dist} -Summary: OpenCL Inference of LLaMA model in pure C/C++ -License: MIT -Source0: https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz -BuildRequires: coreutils make gcc-c++ git mesa-libOpenCL-devel -URL: https://github.com/ggerganov/llama.cpp - -%define debug_package %{nil} -%define source_date_epoch_from_changelog 0 - -%description -CPU inference for Meta's Lllama2 models using default options. - -%prep -%setup -n llama.cpp-master - -%build -make -j LLAMA_CLBLAST=1 - -%install -mkdir -p %{buildroot}%{_bindir}/ -cp -p main %{buildroot}%{_bindir}/llamacppclblast -cp -p server %{buildroot}%{_bindir}/llamacppclblastserver -cp -p simple %{buildroot}%{_bindir}/llamacppclblastsimple - -%clean -rm -rf %{buildroot} -rm -rf %{_builddir}/* - -%files -%{_bindir}/llamacppclblast -%{_bindir}/llamacppclblastserver -%{_bindir}/llamacppclblastsimple - -%pre - -%post - -%preun -%postun - -%changelog diff --git a/.devops/lamma-cpp-cublas.srpm.spec b/.devops/lamma-cpp-cublas.srpm.spec deleted file mode 100644 index 75d32fbe..00000000 --- a/.devops/lamma-cpp-cublas.srpm.spec +++ /dev/null @@ -1,59 +0,0 @@ -# SRPM for building from source and packaging an RPM for RPM-based distros. -# https://fedoraproject.org/wiki/How_to_create_an_RPM_package -# Built and maintained by John Boero - boeroboy@gmail.com -# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal - -# Notes for llama.cpp: -# 1. Tags are currently based on hash - which will not sort asciibetically. -# We need to declare standard versioning if people want to sort latest releases. -# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies. -# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed. -# Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo -# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries. -# It is up to the user to install the correct vendor-specific support. - -Name: llama.cpp-cublas -Version: master -Release: 1%{?dist} -Summary: CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL) -License: MIT -Source0: https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz -BuildRequires: coreutils make gcc-c++ git cuda-toolkit -Requires: cuda-toolkit -URL: https://github.com/ggerganov/llama.cpp - -%define debug_package %{nil} -%define source_date_epoch_from_changelog 0 - -%description -CPU inference for Meta's Lllama2 models using default options. - -%prep -%setup -n llama.cpp-master - -%build -make -j LLAMA_CUBLAS=1 - -%install -mkdir -p %{buildroot}%{_bindir}/ -cp -p main %{buildroot}%{_bindir}/llamacppcublas -cp -p server %{buildroot}%{_bindir}/llamacppcublasserver -cp -p simple %{buildroot}%{_bindir}/llamacppcublassimple - -%clean -rm -rf %{buildroot} -rm -rf %{_builddir}/* - -%files -%{_bindir}/llamacppcublas -%{_bindir}/llamacppcublasserver -%{_bindir}/llamacppcublassimple - -%pre - -%post - -%preun -%postun - -%changelog diff --git a/.devops/llama-cpp-clblast.srpm.spec b/.devops/llama-cpp-clblast.srpm.spec new file mode 100644 index 00000000..076f2969 --- /dev/null +++ b/.devops/llama-cpp-clblast.srpm.spec @@ -0,0 +1,84 @@ +# SRPM for building from source and packaging an RPM for RPM-based distros. +# https://fedoraproject.org/wiki/How_to_create_an_RPM_package +# Built and maintained by John Boero - boeroboy@gmail.com +# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal + +# Notes for llama.cpp: +# 1. Tags are currently based on hash - which will not sort asciibetically. +# We need to declare standard versioning if people want to sort latest releases. +# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies. +# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed. +# Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo +# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries. +# It is up to the user to install the correct vendor-specific support. + +Name: llama.cpp-clblast +Version: %( date "+%%Y%%m%%d" ) +Release: 1%{?dist} +Summary: OpenCL Inference of LLaMA model in C/C++ +License: MIT +Source0: https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz +BuildRequires: coreutils make gcc-c++ git mesa-libOpenCL-devel clblast-devel +Requires: clblast +URL: https://github.com/ggerganov/llama.cpp + +%define debug_package %{nil} +%define source_date_epoch_from_changelog 0 + +%description +CPU inference for Meta's Lllama2 models using default options. + +%prep +%setup -n llama.cpp-master + +%build +make -j LLAMA_CLBLAST=1 + +%install +mkdir -p %{buildroot}%{_bindir}/ +cp -p main %{buildroot}%{_bindir}/llamaclblast +cp -p server %{buildroot}%{_bindir}/llamaclblastserver +cp -p simple %{buildroot}%{_bindir}/llamaclblastsimple + +mkdir -p %{buildroot}/usr/lib/systemd/system +%{__cat} < %{buildroot}/usr/lib/systemd/system/llamaclblast.service +[Unit] +Description=Llama.cpp server, CPU only (no GPU support in this build). +After=syslog.target network.target local-fs.target remote-fs.target nss-lookup.target + +[Service] +Type=simple +EnvironmentFile=/etc/sysconfig/llama +ExecStart=/usr/bin/llamaclblastserver $LLAMA_ARGS +ExecReload=/bin/kill -s HUP $MAINPID +Restart=never + +[Install] +WantedBy=default.target +EOF + +mkdir -p %{buildroot}/etc/sysconfig +%{__cat} < %{buildroot}/etc/sysconfig/llama +LLAMA_ARGS="-m /opt/llama2/ggml-model-f32.bin" +EOF + +%clean +rm -rf %{buildroot} +rm -rf %{_builddir}/* + +%files +%{_bindir}/llamaclblast +%{_bindir}/llamaclblastserver +%{_bindir}/llamaclblastsimple +/usr/lib/systemd/system/llamaclblast.service +%config /etc/sysconfig/llama + + +%pre + +%post + +%preun +%postun + +%changelog diff --git a/.devops/llama-cpp-cublas.srpm.spec b/.devops/llama-cpp-cublas.srpm.spec new file mode 100644 index 00000000..f847ebb1 --- /dev/null +++ b/.devops/llama-cpp-cublas.srpm.spec @@ -0,0 +1,83 @@ +# SRPM for building from source and packaging an RPM for RPM-based distros. +# https://fedoraproject.org/wiki/How_to_create_an_RPM_package +# Built and maintained by John Boero - boeroboy@gmail.com +# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal + +# Notes for llama.cpp: +# 1. Tags are currently based on hash - which will not sort asciibetically. +# We need to declare standard versioning if people want to sort latest releases. +# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies. +# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed. +# Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo +# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries. +# It is up to the user to install the correct vendor-specific support. + +Name: llama.cpp-cublas +Version: %( date "+%%Y%%m%%d" ) +Release: 1%{?dist} +Summary: CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL) +License: MIT +Source0: https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz +BuildRequires: coreutils make gcc-c++ git cuda-toolkit +Requires: cuda-toolkit +URL: https://github.com/ggerganov/llama.cpp + +%define debug_package %{nil} +%define source_date_epoch_from_changelog 0 + +%description +CPU inference for Meta's Lllama2 models using default options. + +%prep +%setup -n llama.cpp-master + +%build +make -j LLAMA_CUBLAS=1 + +%install +mkdir -p %{buildroot}%{_bindir}/ +cp -p main %{buildroot}%{_bindir}/llamacppcublas +cp -p server %{buildroot}%{_bindir}/llamacppcublasserver +cp -p simple %{buildroot}%{_bindir}/llamacppcublassimple + +mkdir -p %{buildroot}/usr/lib/systemd/system +%{__cat} < %{buildroot}/usr/lib/systemd/system/llamacublas.service +[Unit] +Description=Llama.cpp server, CPU only (no GPU support in this build). +After=syslog.target network.target local-fs.target remote-fs.target nss-lookup.target + +[Service] +Type=simple +EnvironmentFile=/etc/sysconfig/llama +ExecStart=/usr/bin/llamacppcublasserver $LLAMA_ARGS +ExecReload=/bin/kill -s HUP $MAINPID +Restart=never + +[Install] +WantedBy=default.target +EOF + +mkdir -p %{buildroot}/etc/sysconfig +%{__cat} < %{buildroot}/etc/sysconfig/llama +LLAMA_ARGS="-m /opt/llama2/ggml-model-f32.bin" +EOF + +%clean +rm -rf %{buildroot} +rm -rf %{_builddir}/* + +%files +%{_bindir}/llamacppcublas +%{_bindir}/llamacppcublasserver +%{_bindir}/llamacppcublassimple +/usr/lib/systemd/system/llamacublas.service +%config /etc/sysconfig/llama + +%pre + +%post + +%preun +%postun + +%changelog diff --git a/.devops/llama-cpp.srpm.spec b/.devops/llama-cpp.srpm.spec index c65251a5..446213d6 100644 --- a/.devops/llama-cpp.srpm.spec +++ b/.devops/llama-cpp.srpm.spec @@ -6,6 +6,7 @@ # Notes for llama.cpp: # 1. Tags are currently based on hash - which will not sort asciibetically. # We need to declare standard versioning if people want to sort latest releases. +# In the meantime, YYYYMMDD format will be used. # 2. Builds for CUDA/OpenCL support are separate, with different depenedencies. # 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed. # Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo @@ -13,12 +14,13 @@ # It is up to the user to install the correct vendor-specific support. Name: llama.cpp -Version: master +Version: %( date "+%%Y%%m%%d" ) Release: 1%{?dist} Summary: CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL) License: MIT Source0: https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz -BuildRequires: coreutils make gcc-c++ git +BuildRequires: coreutils make gcc-c++ git libstdc++-devel +Requires: libstdc++ URL: https://github.com/ggerganov/llama.cpp %define debug_package %{nil} @@ -26,27 +28,52 @@ URL: https://github.com/ggerganov/llama.cpp %description CPU inference for Meta's Lllama2 models using default options. +Models are not included in this package and must be downloaded separately. %prep -%autosetup +%setup -n llama.cpp-master %build make -j %install mkdir -p %{buildroot}%{_bindir}/ -cp -p main %{buildroot}%{_bindir}/llamacpp -cp -p server %{buildroot}%{_bindir}/llamacppserver -cp -p simple %{buildroot}%{_bindir}/llamacppsimple +cp -p main %{buildroot}%{_bindir}/llama +cp -p server %{buildroot}%{_bindir}/llamaserver +cp -p simple %{buildroot}%{_bindir}/llamasimple + +mkdir -p %{buildroot}/usr/lib/systemd/system +%{__cat} < %{buildroot}/usr/lib/systemd/system/llama.service +[Unit] +Description=Llama.cpp server, CPU only (no GPU support in this build). +After=syslog.target network.target local-fs.target remote-fs.target nss-lookup.target + +[Service] +Type=simple +EnvironmentFile=/etc/sysconfig/llama +ExecStart=/usr/bin/llamaserver $LLAMA_ARGS +ExecReload=/bin/kill -s HUP $MAINPID +Restart=never + +[Install] +WantedBy=default.target +EOF + +mkdir -p %{buildroot}/etc/sysconfig +%{__cat} < %{buildroot}/etc/sysconfig/llama +LLAMA_ARGS="-m /opt/llama2/ggml-model-f32.bin" +EOF %clean rm -rf %{buildroot} rm -rf %{_builddir}/* %files -%{_bindir}/llamacpp -%{_bindir}/llamacppserver -%{_bindir}/llamacppsimple +%{_bindir}/llama +%{_bindir}/llamaserver +%{_bindir}/llamasimple +/usr/lib/systemd/system/llama.service +%config /etc/sysconfig/llama %pre