ggml-python 0.0.32__tar.gz → 0.0.34__tar.gz

{ggml_python-0.0.32 → ggml_python-0.0.34}/.github/workflows/wheels-index.yaml

@@ -38,6 +38,7 @@ jobs:
           ./scripts/releases-to-pep-503.sh index/whl/cu121 '^[v]?[0-9]+\.[0-9]+\.[0-9]+-cu121$'
           ./scripts/releases-to-pep-503.sh index/whl/cu122 '^[v]?[0-9]+\.[0-9]+\.[0-9]+-cu122$'
           ./scripts/releases-to-pep-503.sh index/whl/cu123 '^[v]?[0-9]+\.[0-9]+\.[0-9]+-cu123$'
+          ./scripts/releases-to-pep-503.sh index/whl/metal '^[v]?[0-9]+\.[0-9]+\.[0-9]+-metal$'
       - name: Upload artifact
         uses: actions/upload-pages-artifact@v3
         with:

ggml_python-0.0.34/.github/workflows/wheels-metal.yaml

@@ -0,0 +1,87 @@
+name: Build Wheels (Metal)
+
+on: workflow_dispatch
+
+permissions:
+  contents: write
+
+jobs:
+  define_matrix:
+    name: Define Build Matrix
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+    defaults:
+      run:
+        shell: pwsh
+
+    steps:
+      - name: Define Job Output
+        id: set-matrix
+        run: |
+          $matrix = @{
+              'os' = @('macos-11', 'macos-12', 'macos-13')
+              'pyver' = @('3.10', '3.11', '3.12')
+          }
+
+          $matrixOut = ConvertTo-Json $matrix -Compress
+          Write-Output ('matrix=' + $matrixOut) >> $env:GITHUB_OUTPUT
+
+  build_wheels:
+    name: ${{ matrix.os }} Python ${{ matrix.pyver }}
+    needs: define_matrix
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix: ${{ fromJSON(needs.define_matrix.outputs.matrix) }}
+    env:
+      OSVER: ${{ matrix.os }}
+
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: "recursive"
+
+      - uses: actions/setup-python@v4
+        with:
+          python-version: ${{ matrix.pyver }}
+
+      - name: Install Dependencies
+        run: |
+          python -m pip install build wheel cmake
+
+      - name: Build Wheel
+        run: |
+          XCODE15PATH="/Applications/Xcode_15.0.app/Contents/Developer"
+          XCODE15BINPATH="${XCODE15PATH}/Toolchains/XcodeDefault.xctoolchain/usr/bin"
+          export CMAKE_ARGS="-DGGML_NATIVE=off -DGGML_METAL=on"
+          [[ "$OSVER" == "macos-13" ]] && export CC="${XCODE15BINPATH}/cc" && export CXX="${XCODE15BINPATH}/c++" && export MACOSX_DEPLOYMENT_TARGET="13.0"
+          [[ "$OSVER" == "macos-12" ]] && export MACOSX_DEPLOYMENT_TARGET="12.0"
+          [[ "$OSVER" == "macos-11" ]] && export MACOSX_DEPLOYMENT_TARGET="11.0"
+
+          export CMAKE_OSX_ARCHITECTURES="arm64" && export ARCHFLAGS="-arch arm64"
+          VERBOSE=1 python -m build --wheel
+
+          if [[ "$OSVER" == "macos-13" ]]; then
+            export SDKROOT="${XCODE15PATH}/Platforms/MacOSX.platform/Developer/SDKs/MacOSX14.0.sdk"
+            export MACOSX_DEPLOYMENT_TARGET="14.0"
+            VERBOSE=1 python -m build --wheel
+          fi
+
+          for file in ./dist/*.whl; do cp "$file" "${file/arm64.whl/aarch64.whl}"; done
+
+          export CMAKE_OSX_ARCHITECTURES="x86_64" && export CMAKE_ARGS="-DGGML_NATIVE=off -DGGML_AVX=off -DGGML_AVX2=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_METAL=on" && export ARCHFLAGS="-arch x86_64"
+          VERBOSE=1 python -m build --wheel
+
+          if [[ "$OSVER" == "macos-13" ]]; then
+            export SDKROOT="${XCODE15PATH}/Platforms/MacOSX.platform/Developer/SDKs/MacOSX14.0.sdk"
+            export MACOSX_DEPLOYMENT_TARGET="14.0"
+            VERBOSE=1 python -m build --wheel
+          fi
+
+      - uses: softprops/action-gh-release@v1
+        with:
+          files: dist/*
+          # set release name to <tag>-metal
+          tag_name: ${{ github.ref_name }}-metal
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

{ggml_python-0.0.32 → ggml_python-0.0.34}/CMakeLists.txt

@@ -16,6 +16,16 @@ endif()
 
 set(BUILD_SHARED_LIBS "On")
 set(CMAKE_WINDOWS_EXPORT_ALL_SYMBOLS ON)
+if (APPLE)
+    if (CMAKE_SYSTEM_PROCESSOR MATCHES "arm64")
+        set(GGML_AVX "Off" CACHE BOOL "ggml: enable AVX" FORCE)
+        set(GGML_AVX2 "Off" CACHE BOOL "ggml: enable AVX2" FORCE)
+        set(GGML_FMA "Off" CACHE BOOL "ggml: enable FMA" FORCE)
+        set(GGML_F16C "Off" CACHE BOOL "ggml: enable F16C" FORCE)
+    endif()
+
+    set(GGML_METAL_EMBED_LIBRARY "On" CACHE BOOL "ggml: embed metal library" FORCE)
+endif()
 add_subdirectory(vendor/ggml)
 install(
     TARGETS ggml 
@@ -24,4 +34,8 @@ install(
     RUNTIME DESTINATION ${GGML_PYTHON_INSTALL_DIR}
     FRAMEWORK DESTINATION ${GGML_PYTHON_INSTALL_DIR}
     RESOURCE DESTINATION ${GGML_PYTHON_INSTALL_DIR}
+)
+install(
+    FILES $<TARGET_RUNTIME_DLLS:ggml>
+    DESTINATION ${GGML_PYTHON_INSTALL_DIR}
 )

{ggml_python-0.0.32 → ggml_python-0.0.34}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ggml_python
-Version: 0.0.32
+Version: 0.0.34
 Summary: Python bindings for ggml
 Author-Email: Andrei Betlen <abetlen@gmail.com>
 License: MIT

ggml_python-0.0.34/ggml/__init__.py

@@ -0,0 +1,3 @@
+from .ggml import *
+
+__version__ = "0.0.34"

{ggml_python-0.0.32 → ggml_python-0.0.34}/ggml/ggml.py

@@ -10604,7 +10604,7 @@ GGML_CUDA_MAX_DEVICES = 16
 # // backend API
 # GGML_API GGML_CALL ggml_backend_t ggml_backend_cuda_init(int device);
 @ctypes_function(
-    "ggml_backend_cuda_init", [ctypes.c_int], ggml_backend_t, enabled=GGML_USE_CUDA
+    "ggml_backend_cuda_init", [ctypes.c_int], ggml_backend_t_ctypes, enabled=GGML_USE_CUDA
 )
 def ggml_backend_cuda_init(device: int) -> Optional[ggml_backend_t]:
     ...

{ggml_python-0.0.32 → ggml_python-0.0.34}/tests/test_ggml_metal.py

@@ -15,7 +15,6 @@ run_if_ggml_metal_available = pytest.mark.skipif(
     reason="METAL not available",
 )
 
-@pytest.mark.skip(reason="ggml-metal.metal missing ggml-common.h at runtime")
 @run_if_ggml_metal_available
 def test_metal():
     n_tensors = 1 + 2 # input (x) and weights (a, b)

{ggml_python-0.0.32 → ggml_python-0.0.34}/vendor/ggml/CMakeLists.txt

@@ -47,6 +47,7 @@ option(GGML_HIPBLAS                 "ggml: use hipBLAS"                  OFF)
 option(GGML_CUDA                    "ggml: use CUDA"                     OFF)
 option(GGML_CUBLAS                  "ggml: use CUDA (deprecated)"        OFF)
 option(GGML_METAL                   "ggml: use Metal"                    OFF)
+option(GGML_METAL_EMBED_LIBRARY     "ggml: embed Metal library"          OFF)
 
 option(GGML_CUDA_FORCE_DMMV                 "ggml: use dmmv instead of mmvq CUDA kernels"     OFF)
 option(GGML_CUDA_FORCE_MMQ                  "ggml: use mmq kernels instead of cuBLAS"         OFF)

ggml_python-0.0.34/vendor/ggml/docs/logo/ggml-logo-inverted.svg

@@ -0,0 +1,3 @@
+<svg width="256" height="256" viewBox="0 0 256 256" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path fill-rule="evenodd" clip-rule="evenodd" d="M53.0162 51.834V65.694H37.999V190.306H53.0162V204.166H203.008V190.306H218V65.694H203.008V51.834H53.0162ZM65.6934 72.5983H121.069V107.236H107.235V86.4713H79.5535V141.834H107.235V128H93.4007V114.166H121.056V155.694H65.6934V72.5983ZM134.942 100.306H190.344V134.904H176.471V114.166H148.764V169.529H176.471V155.694H162.624V141.834H190.331V183.402H134.929L134.942 100.306Z" fill="white"/>
+</svg>

ggml_python-0.0.34/vendor/ggml/docs/logo/ggml-logo.svg

@@ -0,0 +1,3 @@
+<svg width="256" height="256" viewBox="0 0 256 256" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path fill-rule="evenodd" clip-rule="evenodd" d="M53.0162 51.834V65.694H37.999V190.306H53.0162V204.166H203.008V190.306H218V65.694H203.008V51.834H53.0162ZM65.6934 72.5983H121.069V107.236H107.235V86.4713H79.5535V141.834H107.235V128H93.4007V114.166H121.056V155.694H65.6934V72.5983ZM134.942 100.306H190.344V134.904H176.471V114.166H148.764V169.529H176.471V155.694H162.624V141.834H190.331V183.402H134.929L134.942 100.306Z" fill="black"/>
+</svg>

{ggml_python-0.0.32 → ggml_python-0.0.34}/vendor/ggml/include/ggml/ggml-backend.h

@@ -137,7 +137,7 @@ extern "C" {
     /*
       Example usage:
 
-        // operations that use tensors allocated in a buffer with USAGE_WEIGHTS will be asigned
+        // operations that use tensors allocated in a buffer with USAGE_WEIGHTS will be assigned
         // preferrably to run on the same backend as the buffer
         ggml_backend_buffer_set_usage(buf_weights, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
 

{ggml_python-0.0.32 → ggml_python-0.0.34}/vendor/ggml/src/CMakeLists.txt

@@ -322,6 +322,78 @@ if (GGML_METAL)
     configure_file(ggml-common.h    ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h    COPYONLY)
     configure_file(ggml-metal.metal ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal COPYONLY)
 
+    if (GGML_METAL_EMBED_LIBRARY)
+        enable_language(ASM)
+        add_compile_definitions(GGML_METAL_EMBED_LIBRARY)
+
+        set(METALLIB_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/ggml-common.h")
+        set(METALLIB_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
+
+        file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/autogenerated")
+
+        # merge ggml-common.h and ggml-metal.metal into a single file
+        set(METALLIB_EMBED_ASM    "${CMAKE_BINARY_DIR}/autogenerated/ggml-metal-embed.s")
+        set(METALLIB_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-metal-embed.metal")
+
+        add_custom_command(
+            OUTPUT ${METALLIB_EMBED_ASM}
+            COMMAND echo "Embedding Metal library"
+            COMMAND sed -e '/\#include \"ggml-common.h\"/r ${METALLIB_COMMON}' -e '/\#include \"ggml-common.h\"/d' < ${METALLIB_SOURCE} > ${METALLIB_SOURCE_EMBED}
+            COMMAND echo ".section __DATA,__ggml_metallib"          >  ${METALLIB_EMBED_ASM}
+            COMMAND echo ".globl _ggml_metallib_start"              >> ${METALLIB_EMBED_ASM}
+            COMMAND echo "_ggml_metallib_start:"                    >> ${METALLIB_EMBED_ASM}
+            COMMAND echo ".incbin \\\"${METALLIB_SOURCE_EMBED}\\\"" >> ${METALLIB_EMBED_ASM}
+            COMMAND echo ".globl _ggml_metallib_end"                >> ${METALLIB_EMBED_ASM}
+            COMMAND echo "_ggml_metallib_end:"                      >> ${METALLIB_EMBED_ASM}
+            DEPENDS ggml-metal.metal ggml-common.h
+            COMMENT "Generate assembly for embedded Metal library"
+        )
+
+        set(GGML_METAL_SOURCES ${GGML_METAL_SOURCES} ${METALLIB_EMBED_ASM})
+    else()
+        if (GGML_METAL_SHADER_DEBUG)
+            # custom command to do the following:
+            #   xcrun -sdk macosx metal    -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
+            #   xcrun -sdk macosx metallib                   ggml-metal.air   -o default.metallib
+            #
+            # note: this is the only way I found to disable fast-math in Metal. it's ugly, but at least it works
+            #       disabling fast math is needed in order to pass tests/test-backend-ops
+            # note: adding -fno-inline fixes the tests when using MTL_SHADER_VALIDATION=1
+            # note: unfortunately, we have to call it default.metallib instead of ggml.metallib
+            #       ref: https://github.com/ggerganov/whisper.cpp/issues/1720
+            set(XC_FLAGS -fno-fast-math -fno-inline -g)
+        else()
+            set(XC_FLAGS -O3)
+        endif()
+
+        # Append macOS metal versioning flags
+        if (GGML_METAL_MACOSX_VERSION_MIN)
+            message(STATUS "Adding -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN} flag to metal compilation")
+            list(APPEND XC_FLAGS -mmacosx-version-min=${GGML_METAL_MACOSX_VERSION_MIN})
+        endif()
+        if (GGML_METAL_STD)
+            message(STATUS "Adding -std=${GGML_METAL_STD} flag to metal compilation")
+            list(APPEND XC_FLAGS -std=${GGML_METAL_STD})
+        endif()
+
+        add_custom_command(
+            OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
+            COMMAND xcrun -sdk macosx metal    ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
+            COMMAND xcrun -sdk macosx metallib                ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air   -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal
+            DEPENDS ggml-metal.metal ggml-common.h
+            COMMENT "Compiling Metal kernels"
+            )
+
+        add_custom_target(
+            ggml-metal ALL
+            DEPENDS ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
+            )
+    endif() # GGML_METAL_EMBED_LIBRARY
+
+
     set(GGML_EXTRA_LIBS ${GGML_EXTRA_LIBS}
         ${FOUNDATION_LIBRARY}
         ${METAL_FRAMEWORK}

ggml_python-0.0.32/ggml/__init__.py

@@ -1,3 +0,0 @@
-from .ggml import *
-
-__version__ = "0.0.32"