From 3e8ff47af620a31e0810c58a41e4b089145982ef Mon Sep 17 00:00:00 2001 From: JohnnyB Date: Mon, 28 Aug 2023 07:31:24 +0100 Subject: [PATCH] devops : added systemd units and set versioning to use date. (#2835) * Corrections and systemd units * Missing dependency clblast --- ....srpm.spec => llama-cpp-clblast.srpm.spec} | 44 ++++++++++++++---- ...s.srpm.spec => llama-cpp-cublas.srpm.spec} | 26 ++++++++++- .devops/llama-cpp.srpm.spec | 45 +++++++++++++++---- 3 files changed, 96 insertions(+), 19 deletions(-) rename .devops/{lamma-cpp-clblast.srpm.spec => llama-cpp-clblast.srpm.spec} (56%) rename .devops/{lamma-cpp-cublas.srpm.spec => llama-cpp-cublas.srpm.spec} (71%) diff --git a/.devops/lamma-cpp-clblast.srpm.spec b/.devops/llama-cpp-clblast.srpm.spec similarity index 56% rename from .devops/lamma-cpp-clblast.srpm.spec rename to .devops/llama-cpp-clblast.srpm.spec index 739c68281..076f29695 100644 --- a/.devops/lamma-cpp-clblast.srpm.spec +++ b/.devops/llama-cpp-clblast.srpm.spec @@ -13,12 +13,13 @@ # It is up to the user to install the correct vendor-specific support. Name: llama.cpp-clblast -Version: master +Version: %( date "+%%Y%%m%%d" ) Release: 1%{?dist} -Summary: OpenCL Inference of LLaMA model in pure C/C++ +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 +BuildRequires: coreutils make gcc-c++ git mesa-libOpenCL-devel clblast-devel +Requires: clblast URL: https://github.com/ggerganov/llama.cpp %define debug_package %{nil} @@ -35,18 +36,43 @@ 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 +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}/llamacppclblast -%{_bindir}/llamacppclblastserver -%{_bindir}/llamacppclblastsimple +%{_bindir}/llamaclblast +%{_bindir}/llamaclblastserver +%{_bindir}/llamaclblastsimple +/usr/lib/systemd/system/llamaclblast.service +%config /etc/sysconfig/llama + %pre diff --git a/.devops/lamma-cpp-cublas.srpm.spec b/.devops/llama-cpp-cublas.srpm.spec similarity index 71% rename from .devops/lamma-cpp-cublas.srpm.spec rename to .devops/llama-cpp-cublas.srpm.spec index 75d32fbe7..f847ebb1e 100644 --- a/.devops/lamma-cpp-cublas.srpm.spec +++ b/.devops/llama-cpp-cublas.srpm.spec @@ -13,7 +13,7 @@ # It is up to the user to install the correct vendor-specific support. Name: llama.cpp-cublas -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 @@ -40,6 +40,28 @@ 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}/* @@ -48,6 +70,8 @@ rm -rf %{_builddir}/* %{_bindir}/llamacppcublas %{_bindir}/llamacppcublasserver %{_bindir}/llamacppcublassimple +/usr/lib/systemd/system/llamacublas.service +%config /etc/sysconfig/llama %pre diff --git a/.devops/llama-cpp.srpm.spec b/.devops/llama-cpp.srpm.spec index c65251a5a..446213d69 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