Date: Tue, 6 Aug 2024 10:04:46 GMT From: Yuri Victorovich <yuri@FreeBSD.org> To: ports-committers@FreeBSD.org, dev-commits-ports-all@FreeBSD.org, dev-commits-ports-main@FreeBSD.org Subject: git: 1e343f9e9180 - main - misc/ollama: New port: Run Llama 2, Mistral, and other large language models Message-ID: <202408061004.476A4ka5013495@gitrepo.freebsd.org>
next in thread | raw e-mail | index | archive | help
The branch main has been updated by yuri: URL: https://cgit.FreeBSD.org/ports/commit/?id=1e343f9e9180ff7df20b884b67649ba7d6edc413 commit 1e343f9e9180ff7df20b884b67649ba7d6edc413 Author: Yuri Victorovich <yuri@FreeBSD.org> AuthorDate: 2024-08-06 10:04:03 +0000 Commit: Yuri Victorovich <yuri@FreeBSD.org> CommitDate: 2024-08-06 10:04:44 +0000 misc/ollama: New port: Run Llama 2, Mistral, and other large language models --- misc/Makefile | 1 + misc/ollama/Makefile | 34 +++ misc/ollama/distinfo | 7 + misc/ollama/files/patch-FreeBSD-compatibility | 266 +++++++++++++++++++++ .../ollama/files/patch-llm_generate_gen__common.sh | 35 +++ misc/ollama/pkg-descr | 3 + 6 files changed, 346 insertions(+) diff --git a/misc/Makefile b/misc/Makefile index 10ad9ff1e43c..c7728cbd8c79 100644 --- a/misc/Makefile +++ b/misc/Makefile @@ -287,6 +287,7 @@ SUBDIR += numchar SUBDIR += nut SUBDIR += ola + SUBDIR += ollama SUBDIR += ompl SUBDIR += ondir SUBDIR += onnx diff --git a/misc/ollama/Makefile b/misc/ollama/Makefile new file mode 100644 index 000000000000..b097932499b4 --- /dev/null +++ b/misc/ollama/Makefile @@ -0,0 +1,34 @@ +PORTNAME= ollama +DISTVERSIONPREFIX= v +DISTVERSION= 0.3.4 +CATEGORIES= misc # machine-learning + +MAINTAINER= yuri@FreeBSD.org +COMMENT= Run Llama 2, Mistral, and other large language models +WWW= https://ollama.com/ + +LICENSE= MIT +LICENSE_FILE= ${WRKSRC}/LICENSE + +BUILD_DEPENDS= bash:shells/bash \ + cmake:devel/cmake-core \ + vulkan-headers>0:graphics/vulkan-headers +LIB_DEPENDS= libvulkan.so:graphics/vulkan-loader + +USES= go:1.22,modules pkgconfig + +GO_MODULE= github.com/${PORTNAME}/${PORTNAME} +GO_TARGET= . + +USE_GITHUB= nodefault +GH_TUPLE= ggerganov:llama.cpp:6eeaeba:llama_cpp/llm/llama.cpp + +PLIST_FILES= bin/${PORTNAME} + +pre-build: + @${CP} ${WRKSRC}/app/store/store_linux.go ${WRKSRC}/app/store/store_bsd.go + @cd ${GO_WRKSRC} && \ + ${SETENVI} ${WRK_ENV} ${MAKE_ENV} ${GO_ENV} GOMAXPROCS=${MAKE_JOBS_NUMBER} GOPROXY=off ${GO_CMD} generate ${GO_BUILDFLAGS} \ + ./... + +.include <bsd.port.mk> diff --git a/misc/ollama/distinfo b/misc/ollama/distinfo new file mode 100644 index 000000000000..c436195aa415 --- /dev/null +++ b/misc/ollama/distinfo @@ -0,0 +1,7 @@ +TIMESTAMP = 1722929347 +SHA256 (go/misc_ollama/ollama-v0.3.4/v0.3.4.mod) = ec7b6ce8fd820503671d269fd43ea5ae7275b1ff095c77d6422ac8540ccb410b +SIZE (go/misc_ollama/ollama-v0.3.4/v0.3.4.mod) = 2992 +SHA256 (go/misc_ollama/ollama-v0.3.4/v0.3.4.zip) = d9bc3d3901272f8fbc9b57ce1af1caa2209127978bb5f72257b6a3fb84161969 +SIZE (go/misc_ollama/ollama-v0.3.4/v0.3.4.zip) = 1832031 +SHA256 (go/misc_ollama/ollama-v0.3.4/ggerganov-llama.cpp-6eeaeba_GH0.tar.gz) = f749f494dc5e672926179524cf28c834151edba5acae18e132c954df76ffdf33 +SIZE (go/misc_ollama/ollama-v0.3.4/ggerganov-llama.cpp-6eeaeba_GH0.tar.gz) = 19003544 diff --git a/misc/ollama/files/patch-FreeBSD-compatibility b/misc/ollama/files/patch-FreeBSD-compatibility new file mode 100644 index 000000000000..02af4cec63c0 --- /dev/null +++ b/misc/ollama/files/patch-FreeBSD-compatibility @@ -0,0 +1,266 @@ +-- patch based on https://github.com/ollama/ollama/issues/1102#issuecomment-2270042340 + +new file mode 100644 +--- gpu/gpu_bsd.go.orig 2024-08-06 07:41:26 UTC ++++ gpu/gpu_bsd.go +@@ -0,0 +1,101 @@ ++//go:build dragonfly || freebsd || netbsd || openbsd ++ ++package gpu ++ ++import "github.com/ollama/ollama/format" ++ ++/* ++#cgo CFLAGS: -I/usr/local/include ++#cgo LDFLAGS: -L/usr/local/lib -lvulkan ++ ++#include <stdbool.h> ++#include <unistd.h> ++#include <vulkan/vulkan.h> ++ ++bool hasVulkanSupport(uint64_t *memSize) { ++ VkInstance instance; ++ ++ VkApplicationInfo appInfo = { VK_STRUCTURE_TYPE_APPLICATION_INFO }; ++ appInfo.pApplicationName = "Ollama"; ++ appInfo.apiVersion = VK_API_VERSION_1_0; ++ ++ VkInstanceCreateInfo createInfo = { VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO }; ++ createInfo.pApplicationInfo = &appInfo; ++ ++ // Create a Vulkan instance ++ if (vkCreateInstance(&createInfo, NULL, &instance) != VK_SUCCESS) ++ return false; ++ ++ // Fetch the first physical Vulkan device. Note that numDevices is overwritten with the number of devices found ++ uint32_t numDevices = 1; ++ VkPhysicalDevice device; ++ vkEnumeratePhysicalDevices(instance, &numDevices, &device); ++ if (numDevices == 0) { ++ vkDestroyInstance(instance, NULL); ++ return false; ++ } ++ ++ // Fetch the memory information for this device. ++ VkPhysicalDeviceMemoryProperties memProperties; ++ vkGetPhysicalDeviceMemoryProperties(device, &memProperties); ++ ++ // Add up all the heaps. ++ VkDeviceSize totalMemory = 0; ++ for (uint32_t i = 0; i < memProperties.memoryHeapCount; ++i) { ++ if (memProperties.memoryHeaps[i].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) { ++ *memSize += memProperties.memoryHeaps[i].size; ++ } ++ } ++ ++ vkDestroyInstance(instance, NULL); ++ return true; ++} ++*/ ++import "C" ++ ++func GetGPUInfo() GpuInfoList { ++ var gpuMem C.uint64_t ++ if C.hasVulkanSupport(&gpuMem) { ++ // Vulkan supported ++ return []GpuInfo{ ++ { ++ Library: "vulkan", ++ ID: "0", ++ MinimumMemory: 512 * format.MebiByte, ++ memInfo: memInfo{ ++ FreeMemory: uint64(gpuMem), ++ TotalMemory: uint64(gpuMem), ++ }, ++ }, ++ } ++ } ++ ++ // CPU fallback ++ cpuMem, _ := GetCPUMem() ++ return []GpuInfo{ ++ { ++ Library: "cpu", ++ memInfo: cpuMem, ++ }, ++ } ++} ++ ++func GetCPUInfo() GpuInfoList { ++ mem, _ := GetCPUMem() ++ return []GpuInfo{ ++ { ++ Library: "cpu", ++ Variant: GetCPUCapability(), ++ memInfo: mem, ++ }, ++ } ++} ++ ++func GetCPUMem() (memInfo, error) { ++ size := C.sysconf(C._SC_PHYS_PAGES) * C.sysconf(C._SC_PAGE_SIZE) ++ return memInfo{TotalMemory: uint64(size)}, nil ++} ++ ++func (l GpuInfoList) GetVisibleDevicesEnv() (string, string) { ++ return "", "" ++} +--- gpu/gpu_test.go.orig 1979-11-30 08:00:00 UTC ++++ gpu/gpu_test.go +@@ -11,7 +11,7 @@ func TestBasicGetGPUInfo(t *testing.T) { + func TestBasicGetGPUInfo(t *testing.T) { + info := GetGPUInfo() + assert.NotEmpty(t, len(info)) +- assert.Contains(t, "cuda rocm cpu metal", info[0].Library) ++ assert.Contains(t, "cuda rocm cpu metal vulkan", info[0].Library) + if info[0].Library != "cpu" { + assert.Greater(t, info[0].TotalMemory, uint64(0)) + assert.Greater(t, info[0].FreeMemory, uint64(0)) +@@ -24,6 +24,8 @@ func TestCPUMemInfo(t *testing.T) { + switch runtime.GOOS { + case "darwin": + t.Skip("CPU memory not populated on darwin") ++ case "dragonfly", "freebsd", "netbsd", "openbsd": ++ t.Skip("CPU memory is not populated on *BSD") + case "linux", "windows": + assert.Greater(t, info.TotalMemory, uint64(0)) + assert.Greater(t, info.FreeMemory, uint64(0)) +--- llm/generate/gen_bsd.sh.orig 2024-08-06 07:41:26 UTC ++++ llm/generate/gen_bsd.sh +@@ -0,0 +1,64 @@ ++#!/bin/sh ++# This script is intended to run inside the go generate ++# working directory must be ./llm/generate/ ++ ++set -ex ++set -o pipefail ++echo "Starting BSD generate script" ++. $(dirname $0)/gen_common.sh ++init_vars ++#git_module_setup ++apply_patches ++ ++COMMON_BSD_DEFS="-DCMAKE_SYSTEM_NAME=$(uname -s)" ++CMAKE_TARGETS="--target llama --target ggml" ++ ++case "${GOARCH}" in ++ "amd64") ++ COMMON_CPU_DEFS="${COMMON_BSD_DEFS} -DCMAKE_SYSTEM_PROCESSOR=${ARCH}" ++ ++ # Static build for linking into the Go binary ++ init_vars ++ CMAKE_DEFS="${COMMON_CPU_DEFS} -DBUILD_SHARED_LIBS=off -DLLAMA_ACCELERATE=off -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}" ++ BUILD_DIR="../build/bsd/${ARCH}_static" ++ echo "Building static library" ++ build ++ ++ init_vars ++ CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}" ++ BUILD_DIR="../build/bsd/${ARCH}/cpu" ++ echo "Building LCD CPU" ++ build ++ compress ++ ++ init_vars ++ CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}" ++ BUILD_DIR="../build/bsd/${ARCH}/cpu_avx" ++ echo "Building AVX CPU" ++ build ++ compress ++ ++ init_vars ++ CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=on -DLLAMA_AVX2=on -DLLAMA_AVX512=off -DLLAMA_FMA=on -DLLAMA_F16C=on ${CMAKE_DEFS}" ++ BUILD_DIR="../build/bsd/${ARCH}/cpu_avx2" ++ echo "Building AVX2 CPU" ++ build ++ compress ++ ++ init_vars ++ CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_VULKAN=on ${CMAKE_DEFS}" ++ BUILD_DIR="../build/bsd/${ARCH}/vulkan" ++ echo "Building Vulkan GPU" ++ build ++ compress ++ ;; ++ ++ *) ++ echo "GOARCH must be set" ++ echo "this script is meant to be run from within go generate" ++ exit 1 ++ ;; ++esac ++ ++cleanup ++echo "go generate completed. LLM runners: $(cd ${BUILD_DIR}/..; echo *)" +--- llm/generate/generate_bsd.go.orig 2024-08-06 07:41:26 UTC ++++ llm/generate/generate_bsd.go +@@ -0,0 +1,5 @@ ++//go:build dragonfly || freebsd || netbsd || openbsd ++ ++package generate ++ ++//go:generate bash ./gen_bsd.sh +--- llm/llm.go.orig 1979-11-30 08:00:00 UTC ++++ llm/llm.go +@@ -8,6 +8,10 @@ package llm + // #cgo windows,arm64 LDFLAGS: -static-libstdc++ -static-libgcc -static -L${SRCDIR}/build/windows/arm64_static -L${SRCDIR}/build/windows/arm64_static/src -L${SRCDIR}/build/windows/arm64_static/ggml/src + // #cgo linux,amd64 LDFLAGS: -L${SRCDIR}/build/linux/x86_64_static -L${SRCDIR}/build/linux/x86_64_static/src -L${SRCDIR}/build/linux/x86_64_static/ggml/src + // #cgo linux,arm64 LDFLAGS: -L${SRCDIR}/build/linux/arm64_static -L${SRCDIR}/build/linux/arm64_static/src -L${SRCDIR}/build/linux/arm64_static/ggml/src ++// #cgo dragonfly,amd64 LDFLAGS: ${SRCDIR}/build/bsd/x86_64_static/src/libllama.a -lstdc++ -lm ++// #cgo freebsd,amd64 LDFLAGS: -L${SRCDIR}/build/bsd/x86_64_static/src -lllama -L${SRCDIR}/build/bsd/x86_64_static/ggml/src -lggml -lstdc++ -lm -lomp ++// #cgo netbsd,amd64 LDFLAGS: ${SRCDIR}/build/bsd/x86_64_static/src/libllama.a -lstdc++ -lm ++// #cgo openbsd,amd64 LDFLAGS: ${SRCDIR}/build/bsd/x86_64_static/src/libllama.a -lstdc++ -lm + // #include <stdlib.h> + // #include "llama.h" + import "C" +--- llm/llm_bsd.go.orig 2024-08-06 07:41:26 UTC ++++ llm/llm_bsd.go +@@ -0,0 +1,13 @@ ++//go:build dragonfly || freebsd || netbsd || openbsd ++ ++package llm ++ ++import ( ++ "embed" ++ "syscall" ++) ++ ++//go:embed build/bsd/*/*/bin/* ++var libEmbed embed.FS ++ ++var LlamaServerSysProcAttr = &syscall.SysProcAttr{} +--- scripts/build_bsd.sh.orig 2024-08-06 07:41:26 UTC ++++ scripts/build_bsd.sh +@@ -0,0 +1,27 @@ ++#!/bin/sh ++ ++set -e ++ ++case "$(uname -s)" in ++ DragonFly) ++ ;; ++ FreeBSD) ++ ;; ++ NetBSD) ++ ;; ++ OpenBSD) ++ ;; ++ *) ++ echo "$(uname -s) is not supported" ++ exit 1 ++ ;; ++esac ++ ++export VERSION=${VERSION:-$(git describe --tags --first-parent --abbrev=7 --long --dirty --always | sed -e "s/^v//g")} ++export GOFLAGS="'-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=$VERSION\" \"-X=github.com/ollama/ollama/server.mode=release\"'" ++ ++mkdir -p dist ++rm -rf llm/llama.cpp/build ++ ++go generate ./... ++CGO_ENABLED=1 go build -trimpath -o dist/ollama-bsd +--- scripts/build_freebsd.sh.orig 2024-08-06 07:41:26 UTC ++++ scripts/build_freebsd.sh +@@ -0,0 +1 @@ ++build_bsd.sh +\ No newline at end of file diff --git a/misc/ollama/files/patch-llm_generate_gen__common.sh b/misc/ollama/files/patch-llm_generate_gen__common.sh new file mode 100644 index 000000000000..6033d77dc663 --- /dev/null +++ b/misc/ollama/files/patch-llm_generate_gen__common.sh @@ -0,0 +1,35 @@ +--- llm/generate/gen_common.sh.orig 1979-11-30 08:00:00 UTC ++++ llm/generate/gen_common.sh +@@ -68,13 +68,8 @@ apply_patches() { + if [ -n "$(ls -A ../patches/*.diff)" ]; then + # apply temporary patches until fix is upstream + for patch in ../patches/*.diff; do +- for file in $(grep "^+++ " ${patch} | cut -f2 -d' ' | cut -f2- -d/); do +- (cd ${LLAMACPP_DIR}; git checkout ${file}) +- done ++ (cd ${LLAMACPP_DIR} && patch -B '' -p1 < ${patch}) + done +- for patch in ../patches/*.diff; do +- (cd ${LLAMACPP_DIR} && git apply ${patch}) +- done + fi + } + +@@ -106,14 +101,8 @@ compress() { + } + + # Keep the local tree clean after we're done with the build +-cleanup() { +- (cd ${LLAMACPP_DIR}/ && git checkout CMakeLists.txt) + +- if [ -n "$(ls -A ../patches/*.diff)" ]; then +- for patch in ../patches/*.diff; do +- for file in $(grep "^+++ " ${patch} | cut -f2 -d' ' | cut -f2- -d/); do +- (cd ${LLAMACPP_DIR}; git checkout ${file}) +- done +- done +- fi ++cleanup() { ++ echo "cleanup done" + } ++ diff --git a/misc/ollama/pkg-descr b/misc/ollama/pkg-descr new file mode 100644 index 000000000000..96a69944f608 --- /dev/null +++ b/misc/ollama/pkg-descr @@ -0,0 +1,3 @@ +Ollama allows to get up and running with large language models. + +Ollama supports a list of models available on ollama.com/library.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202408061004.476A4ka5013495>