mirror of
https://github.com/ollama/ollama.git
synced 2026-01-12 00:06:57 +08:00
33ee7168ba
* WIP - MLX backend with gemma3 * MLX: add cmake and go tag build toggles To build the new MLX backend code: cmake --preset MLX cmake --build --preset MLX --parallel cmake --install build --component MLX go build -tags mlx . Note: the main.go entrypoint for the MLX engine will change in a follow up commit. * add experimental image generation runtime * add experimental image generation runtime * MLX: wire up cuda build for linux * MLX: get dependencies correct and dedup This is still too large for a unified github artifact, but is now "correct" for the mlx_cuda_v13 directory. * fix relative link bug in dedup * Add darwin build and readme * add go build tag for mlx dependent code and wire up build_darwin.sh * lint cleanup * macos: build mlx for x86 This will be CPU only. * cuda build instructions and fix drift from mlx bump * stale comment * Delete agent helper doc * Clean up readme.md * Revise README for tokenizer clarity and details Updated README to clarify tokenizer functionality and removed correctness section. --------- Co-authored-by: jmorganca <jmorganca@gmail.com>
210 lines
8.1 KiB
CMake
210 lines
8.1 KiB
CMake
cmake_minimum_required(VERSION 3.21)
|
|
|
|
project(Ollama C CXX)
|
|
|
|
# Handle cross-compilation on macOS: when CMAKE_OSX_ARCHITECTURES is set to a
|
|
# single architecture different from the host, override CMAKE_SYSTEM_PROCESSOR
|
|
# to match. This is necessary because CMAKE_SYSTEM_PROCESSOR defaults to the
|
|
# host architecture, but downstream projects (like MLX) use it to detect the
|
|
# target architecture.
|
|
if(CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES ";")
|
|
# Single architecture specified
|
|
if(CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" AND NOT CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64")
|
|
message(STATUS "Cross-compiling for x86_64: overriding CMAKE_SYSTEM_PROCESSOR from ${CMAKE_SYSTEM_PROCESSOR} to x86_64")
|
|
set(CMAKE_SYSTEM_PROCESSOR "x86_64")
|
|
elseif(CMAKE_OSX_ARCHITECTURES STREQUAL "arm64" AND NOT CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
|
|
message(STATUS "Cross-compiling for arm64: overriding CMAKE_SYSTEM_PROCESSOR from ${CMAKE_SYSTEM_PROCESSOR} to arm64")
|
|
set(CMAKE_SYSTEM_PROCESSOR "arm64")
|
|
endif()
|
|
endif()
|
|
|
|
include(CheckLanguage)
|
|
include(GNUInstallDirs)
|
|
|
|
find_package(Threads REQUIRED)
|
|
|
|
set(CMAKE_BUILD_TYPE Release)
|
|
set(BUILD_SHARED_LIBS ON)
|
|
|
|
set(CMAKE_CXX_STANDARD 17)
|
|
set(CMAKE_CXX_STANDARD_REQUIRED ON)
|
|
set(CMAKE_CXX_EXTENSIONS ON) # Recent versions of MLX Requires gnu++17 extensions to compile properly
|
|
|
|
set(GGML_BUILD ON)
|
|
set(GGML_SHARED ON)
|
|
set(GGML_CCACHE ON)
|
|
set(GGML_BACKEND_DL ON)
|
|
set(GGML_BACKEND_SHARED ON)
|
|
set(GGML_SCHED_MAX_COPIES 4)
|
|
|
|
set(GGML_LLAMAFILE ON)
|
|
set(GGML_CUDA_PEER_MAX_BATCH_SIZE 128)
|
|
set(GGML_CUDA_GRAPHS ON)
|
|
set(GGML_CUDA_FA ON)
|
|
set(GGML_CUDA_COMPRESSION_MODE default)
|
|
|
|
if((CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
|
|
OR (NOT CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_SYSTEM_PROCESSOR MATCHES "arm|aarch64|ARM64|ARMv[0-9]+"))
|
|
set(GGML_CPU_ALL_VARIANTS ON)
|
|
endif()
|
|
|
|
if (CMAKE_OSX_ARCHITECTURES MATCHES "x86_64")
|
|
set(CMAKE_BUILD_RPATH "@loader_path")
|
|
set(CMAKE_INSTALL_RPATH "@loader_path")
|
|
endif()
|
|
|
|
set(OLLAMA_BUILD_DIR ${CMAKE_BINARY_DIR}/lib/ollama)
|
|
set(OLLAMA_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/lib/ollama/${OLLAMA_RUNNER_DIR})
|
|
|
|
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${OLLAMA_BUILD_DIR})
|
|
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY_DEBUG ${OLLAMA_BUILD_DIR})
|
|
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY_RELEASE ${OLLAMA_BUILD_DIR})
|
|
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${OLLAMA_BUILD_DIR})
|
|
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY_DEBUG ${OLLAMA_BUILD_DIR})
|
|
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY_RELEASE ${OLLAMA_BUILD_DIR})
|
|
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src)
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/include)
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-cpu)
|
|
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-cpu/amx)
|
|
|
|
add_compile_definitions(NDEBUG GGML_VERSION=0x0 GGML_COMMIT=0x0)
|
|
|
|
# Define GGML version variables for shared library SOVERSION
|
|
# These are required by ggml/src/CMakeLists.txt for proper library versioning
|
|
set(GGML_VERSION_MAJOR 0)
|
|
set(GGML_VERSION_MINOR 0)
|
|
set(GGML_VERSION_PATCH 0)
|
|
set(GGML_VERSION "${GGML_VERSION_MAJOR}.${GGML_VERSION_MINOR}.${GGML_VERSION_PATCH}")
|
|
|
|
set(GGML_CPU ON)
|
|
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src)
|
|
set_property(TARGET ggml PROPERTY EXCLUDE_FROM_ALL TRUE)
|
|
|
|
get_target_property(CPU_VARIANTS ggml-cpu MANUALLY_ADDED_DEPENDENCIES)
|
|
if(NOT CPU_VARIANTS)
|
|
set(CPU_VARIANTS "ggml-cpu")
|
|
endif()
|
|
|
|
install(TARGETS ggml-base ${CPU_VARIANTS}
|
|
RUNTIME_DEPENDENCIES
|
|
PRE_EXCLUDE_REGEXES ".*"
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT CPU
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT CPU
|
|
FRAMEWORK DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT CPU
|
|
)
|
|
|
|
check_language(CUDA)
|
|
if(CMAKE_CUDA_COMPILER)
|
|
if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.24" AND NOT CMAKE_CUDA_ARCHITECTURES)
|
|
set(CMAKE_CUDA_ARCHITECTURES "native")
|
|
endif()
|
|
|
|
find_package(CUDAToolkit)
|
|
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-cuda)
|
|
install(TARGETS ggml-cuda
|
|
RUNTIME_DEPENDENCIES
|
|
DIRECTORIES ${CUDAToolkit_BIN_DIR} ${CUDAToolkit_BIN_DIR}/x64 ${CUDAToolkit_LIBRARY_DIR}
|
|
PRE_INCLUDE_REGEXES cublas cublasLt cudart
|
|
PRE_EXCLUDE_REGEXES ".*"
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT CUDA
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT CUDA
|
|
)
|
|
endif()
|
|
|
|
set(WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX "^gfx(908|90a|1200|1201):xnack[+-]$"
|
|
CACHE STRING
|
|
"Regular expression describing AMDGPU_TARGETS not supported on Windows. Override to force building these targets. Default \"^gfx(908|90a|1200|1201):xnack[+-]$\"."
|
|
)
|
|
|
|
check_language(HIP)
|
|
if(CMAKE_HIP_COMPILER)
|
|
set(HIP_PLATFORM "amd")
|
|
|
|
if(NOT AMDGPU_TARGETS)
|
|
find_package(hip REQUIRED)
|
|
list(FILTER AMDGPU_TARGETS INCLUDE REGEX "^gfx(94[012]|101[02]|1030|110[012]|120[01])$")
|
|
endif()
|
|
|
|
if(WIN32 AND WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX)
|
|
list(FILTER AMDGPU_TARGETS EXCLUDE REGEX ${WINDOWS_AMDGPU_TARGETS_EXCLUDE_REGEX})
|
|
endif()
|
|
|
|
if(AMDGPU_TARGETS)
|
|
find_package(hip REQUIRED)
|
|
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-hip)
|
|
|
|
if (WIN32)
|
|
target_compile_definitions(ggml-hip PRIVATE GGML_CUDA_NO_PEER_COPY)
|
|
endif()
|
|
|
|
target_compile_definitions(ggml-hip PRIVATE GGML_HIP_NO_VMM)
|
|
|
|
install(TARGETS ggml-hip
|
|
RUNTIME_DEPENDENCY_SET rocm
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT HIP
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT HIP
|
|
)
|
|
install(RUNTIME_DEPENDENCY_SET rocm
|
|
DIRECTORIES ${HIP_BIN_INSTALL_DIR} ${HIP_LIB_INSTALL_DIR}
|
|
PRE_INCLUDE_REGEXES hipblas rocblas amdhip64 rocsolver amd_comgr hsa-runtime64 rocsparse tinfo rocprofiler-register drm drm_amdgpu numa elf
|
|
PRE_EXCLUDE_REGEXES ".*"
|
|
POST_EXCLUDE_REGEXES "system32"
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT HIP
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT HIP
|
|
)
|
|
|
|
foreach(HIP_LIB_BIN_INSTALL_DIR IN ITEMS ${HIP_BIN_INSTALL_DIR} ${HIP_LIB_INSTALL_DIR})
|
|
if(EXISTS ${HIP_LIB_BIN_INSTALL_DIR}/rocblas)
|
|
install(DIRECTORY ${HIP_LIB_BIN_INSTALL_DIR}/rocblas DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT HIP)
|
|
break()
|
|
endif()
|
|
endforeach()
|
|
endif()
|
|
endif()
|
|
|
|
if(NOT APPLE)
|
|
find_package(Vulkan)
|
|
if(Vulkan_FOUND)
|
|
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/ml/backend/ggml/ggml/src/ggml-vulkan)
|
|
install(TARGETS ggml-vulkan
|
|
RUNTIME_DEPENDENCIES
|
|
PRE_INCLUDE_REGEXES vulkan
|
|
PRE_EXCLUDE_REGEXES ".*"
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT Vulkan
|
|
)
|
|
endif()
|
|
endif()
|
|
|
|
option(MLX_ENGINE "Enable MLX backend" OFF)
|
|
|
|
if(MLX_ENGINE)
|
|
message(STATUS "Setting up MLX (this takes a while...)")
|
|
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/x/ml/backend/mlx)
|
|
|
|
# Find CUDA toolkit if MLX is built with CUDA support
|
|
find_package(CUDAToolkit)
|
|
|
|
install(TARGETS mlx mlxc
|
|
RUNTIME_DEPENDENCIES
|
|
DIRECTORIES ${CUDAToolkit_BIN_DIR} ${CUDAToolkit_BIN_DIR}/x64 ${CUDAToolkit_LIBRARY_DIR}
|
|
PRE_INCLUDE_REGEXES cublas cublasLt cudart nvrtc cudnn nccl
|
|
PRE_EXCLUDE_REGEXES ".*"
|
|
RUNTIME DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
|
|
LIBRARY DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
|
|
FRAMEWORK DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
|
|
)
|
|
|
|
# Manually install cudart and cublas since they might not be picked up as direct dependencies
|
|
if(CUDAToolkit_FOUND)
|
|
file(GLOB CUDART_LIBS
|
|
"${CUDAToolkit_LIBRARY_DIR}/libcudart.so*"
|
|
"${CUDAToolkit_LIBRARY_DIR}/libcublas.so*")
|
|
if(CUDART_LIBS)
|
|
install(FILES ${CUDART_LIBS}
|
|
DESTINATION ${OLLAMA_INSTALL_DIR}
|
|
COMPONENT MLX)
|
|
endif()
|
|
endif()
|
|
endif() |