ci: fix missing Kompute support in python bindings (#1953)

Signed-off-by: Jared Van Bortel <jared@nomic.ai>

.circleci/continue_config.yml

@@ -235,10 +235,8 @@ jobs:
           name: Build
           command: |
             export CMAKE_PREFIX_PATH=~/Qt/6.5.1/gcc_64/lib/cmake
-            mkdir build
-            cd build
-            ~/Qt/Tools/CMake/bin/cmake -DCMAKE_BUILD_TYPE=Release -S ../gpt4all-chat -B .
-            ~/Qt/Tools/CMake/bin/cmake --build . --target all
+            ~/Qt/Tools/CMake/bin/cmake -DCMAKE_BUILD_TYPE=Release -S gpt4all-chat -B build
+            ~/Qt/Tools/CMake/bin/cmake --build build --target all
 
   build-gpt4all-chat-windows:
     machine:
@@ -291,17 +289,15 @@ jobs:
             $Env:INCLUDE = "${Env:INCLUDE};C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\include"
             $Env:INCLUDE = "${Env:INCLUDE};C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.29.30133\ATLMFC\include"
             $Env:VULKAN_SDK = "C:\VulkanSDK\1.3.261.1"
-            mkdir build
-            cd build
             & "C:\Qt\Tools\CMake_64\bin\cmake.exe" `
               "-DCMAKE_GENERATOR:STRING=Ninja" `
               "-DCMAKE_BUILD_TYPE=Release" `
               "-DCMAKE_PREFIX_PATH:PATH=C:\Qt\6.5.1\msvc2019_64" `
               "-DCMAKE_MAKE_PROGRAM:FILEPATH=C:\Qt\Tools\Ninja\ninja.exe" `
               "-DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON" `
-              "-S ..\gpt4all-chat" `
-              "-B ."
-            & "C:\Qt\Tools\Ninja\ninja.exe"
+              "-S gpt4all-chat" `
+              "-B build"
+            & "C:\Qt\Tools\Ninja\ninja.exe" -C build
 
   build-gpt4all-chat-macos:
     macos:
@@ -332,17 +328,15 @@ jobs:
       - run:
           name: Build
           command: |
-            mkdir build
-            cd build
             ~/Qt/Tools/CMake/CMake.app/Contents/bin/cmake \
               -DCMAKE_GENERATOR:STRING=Ninja \
               -DBUILD_UNIVERSAL=ON \
               -DCMAKE_BUILD_TYPE=Release \
               -DCMAKE_PREFIX_PATH:PATH=~/Qt/6.5.1/macos/lib/cmake/Qt6 \
               -DCMAKE_MAKE_PROGRAM:FILEPATH=~/Qt/Tools/Ninja/ninja \
-              -S ../gpt4all-chat \
-              -B .
-            ~/Qt/Tools/CMake/CMake.app/Contents/bin/cmake --build . --target all
+              -S gpt4all-chat \
+              -B build
+            ~/Qt/Tools/CMake/CMake.app/Contents/bin/cmake --build build --target all
   build-ts-docs: 
     docker: 
       - image: cimg/base:stable
@@ -407,13 +401,10 @@ jobs:
       - run:
           name: Build C library
           command: |
-            git submodule init
-            git submodule update
+            git submodule update --init --recursive
             cd gpt4all-backend
-            mkdir build
-            cd build
-            cmake ..
-            cmake --build . --parallel
+            cmake -B build
+            cmake --build build --parallel
       - run:
           name: Build wheel
           command: |
@@ -440,13 +431,10 @@ jobs:
       - run:
           name: Build C library
           command: |
-            git submodule init
-            git submodule update
+            git submodule update --init  # don't use --recursive because macOS doesn't use Kompute
             cd gpt4all-backend
-            mkdir build
-            cd build
-            cmake .. -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64"
-            cmake --build . --parallel
+            cmake -B build -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64"
+            cmake --build build --parallel
       - run:
           name: Build wheel
           command: |
@@ -482,16 +470,13 @@ jobs:
       - run:
           name: Build C library
           command: |
-            git submodule init
-            git submodule update
+            git submodule update --init --recursive
             cd gpt4all-backend
-            mkdir build
-            cd build
             $Env:Path += ";C:\ProgramData\mingw64\mingw64\bin"
             $Env:Path += ";C:\VulkanSDK\1.3.261.1\bin"
             $Env:VULKAN_SDK = "C:\VulkanSDK\1.3.261.1"
-            cmake -G "MinGW Makefiles" .. -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DKOMPUTE_OPT_USE_BUILT_IN_VULKAN_HEADER=OFF
-            cmake --build . --parallel
+            cmake -G "MinGW Makefiles" -B build -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DKOMPUTE_OPT_USE_BUILT_IN_VULKAN_HEADER=OFF
+            cmake --build build --parallel
       - run:
           name: Build wheel
           # TODO: As part of this task, we need to move mingw64 binaries into package.

gpt4all-backend/CMakeLists.txt

@@ -39,10 +39,6 @@ else()
     message(STATUS "Interprocedural optimization support detected")
 endif()
 
-if(NOT APPLE)
-  set(LLAMA_KOMPUTE YES)
-endif()
-
 include(llama.cpp.cmake)
 
 set(BUILD_VARIANTS default avxonly)

gpt4all-backend/llama.cpp.cmake

@@ -71,12 +71,19 @@ option(LLAMA_SANITIZE_UNDEFINED     "llama: enable undefined sanitizer"
 #    option(LLAMA_F16C               "llama: enable F16C"                                    ON)
 #endif()
 
+if (APPLE)
+    set(LLAMA_KOMPUTE_DEFAULT ON)
+else()
+    set(LLAMA_KOMPUTE_DEFAULT OFF)
+endif()
+
 # 3rd party libs
 option(LLAMA_ACCELERATE             "llama: enable Accelerate framework"                    ON)
 option(LLAMA_OPENBLAS               "llama: use OpenBLAS"                                   OFF)
 #option(LLAMA_CUBLAS                 "llama: use cuBLAS"                                     OFF)
 #option(LLAMA_CLBLAST                "llama: use CLBlast"                                    OFF)
 #option(LLAMA_METAL                  "llama: use Metal"                                      OFF)
+option(LLAMA_KOMPUTE                "llama: use Kompute"                                    ${LLAMA_KOMPUTE_DEFAULT})
 set(LLAMA_BLAS_VENDOR "Generic" CACHE STRING "llama: BLAS library vendor")
 set(LLAMA_CUDA_DMMV_X "32" CACHE STRING "llama: x stride for dmmv CUDA kernels")
 set(LLAMA_CUDA_DMMV_Y "1" CACHE STRING  "llama: y block size for dmmv CUDA kernels")
@@ -153,6 +160,11 @@ if (LLAMA_OPENBLAS)
 endif()
 
 if (LLAMA_KOMPUTE)
+    if (NOT EXISTS "${LLAMA_DIR}/kompute/CMakeLists.txt")
+        message(FATAL_ERROR "Kompute not found")
+    endif()
+    message(STATUS "Kompute found")
+
     add_compile_definitions(VULKAN_HPP_DISPATCH_LOADER_DYNAMIC=1)
     find_package(Vulkan COMPONENTS glslc REQUIRED)
     find_program(glslc_executable NAMES glslc HINTS Vulkan::glslc)
@@ -220,91 +232,86 @@ if (LLAMA_KOMPUTE)
       endforeach()
     endfunction()
 
-    if (EXISTS "${LLAMA_DIR}/kompute/CMakeLists.txt")
-        message(STATUS "Kompute found")
-        set(KOMPUTE_OPT_LOG_LEVEL Critical CACHE STRING "Kompute log level")
-        add_subdirectory(${LLAMA_DIR}/kompute)
+    set(KOMPUTE_OPT_LOG_LEVEL Critical CACHE STRING "Kompute log level")
+    add_subdirectory(${LLAMA_DIR}/kompute)
 
-        # Compile our shaders
-        compile_shader(SOURCES
-          kompute-shaders/op_scale.comp
-          kompute-shaders/op_scale_8.comp
-          kompute-shaders/op_add.comp
-          kompute-shaders/op_addrow.comp
-          kompute-shaders/op_mul.comp
-          kompute-shaders/op_silu.comp
-          kompute-shaders/op_relu.comp
-          kompute-shaders/op_gelu.comp
-          kompute-shaders/op_softmax.comp
-          kompute-shaders/op_norm.comp
-          kompute-shaders/op_rmsnorm.comp
-          kompute-shaders/op_diagmask.comp
-          kompute-shaders/op_mul_mat_mat_f32.comp
-          kompute-shaders/op_mul_mat_f16.comp
-          kompute-shaders/op_mul_mat_q8_0.comp
-          kompute-shaders/op_mul_mat_q4_0.comp
-          kompute-shaders/op_mul_mat_q4_1.comp
-          kompute-shaders/op_mul_mat_q6_k.comp
-          kompute-shaders/op_getrows_f16.comp
-          kompute-shaders/op_getrows_q4_0.comp
-          kompute-shaders/op_getrows_q4_1.comp
-          kompute-shaders/op_getrows_q6_k.comp
-          kompute-shaders/op_rope_f16.comp
-          kompute-shaders/op_rope_f32.comp
-          kompute-shaders/op_cpy_f16_f16.comp
-          kompute-shaders/op_cpy_f16_f32.comp
-          kompute-shaders/op_cpy_f32_f16.comp
-          kompute-shaders/op_cpy_f32_f32.comp
-        )
+    # Compile our shaders
+    compile_shader(SOURCES
+      kompute-shaders/op_scale.comp
+      kompute-shaders/op_scale_8.comp
+      kompute-shaders/op_add.comp
+      kompute-shaders/op_addrow.comp
+      kompute-shaders/op_mul.comp
+      kompute-shaders/op_silu.comp
+      kompute-shaders/op_relu.comp
+      kompute-shaders/op_gelu.comp
+      kompute-shaders/op_softmax.comp
+      kompute-shaders/op_norm.comp
+      kompute-shaders/op_rmsnorm.comp
+      kompute-shaders/op_diagmask.comp
+      kompute-shaders/op_mul_mat_mat_f32.comp
+      kompute-shaders/op_mul_mat_f16.comp
+      kompute-shaders/op_mul_mat_q8_0.comp
+      kompute-shaders/op_mul_mat_q4_0.comp
+      kompute-shaders/op_mul_mat_q4_1.comp
+      kompute-shaders/op_mul_mat_q6_k.comp
+      kompute-shaders/op_getrows_f16.comp
+      kompute-shaders/op_getrows_q4_0.comp
+      kompute-shaders/op_getrows_q4_1.comp
+      kompute-shaders/op_getrows_q6_k.comp
+      kompute-shaders/op_rope_f16.comp
+      kompute-shaders/op_rope_f32.comp
+      kompute-shaders/op_cpy_f16_f16.comp
+      kompute-shaders/op_cpy_f16_f32.comp
+      kompute-shaders/op_cpy_f32_f16.comp
+      kompute-shaders/op_cpy_f32_f32.comp
+    )
 
-        # Create a custom target for our generated shaders
-        add_custom_target(generated_shaders DEPENDS
-          shaderop_scale.h
-          shaderop_scale_8.h
-          shaderop_add.h
-          shaderop_addrow.h
-          shaderop_mul.h
-          shaderop_silu.h
-          shaderop_relu.h
-          shaderop_gelu.h
-          shaderop_softmax.h
-          shaderop_norm.h
-          shaderop_rmsnorm.h
-          shaderop_diagmask.h
-          shaderop_mul_mat_mat_f32.h
-          shaderop_mul_mat_f16.h
-          shaderop_mul_mat_q8_0.h
-          shaderop_mul_mat_q4_0.h
-          shaderop_mul_mat_q4_1.h
-          shaderop_mul_mat_q6_k.h
-          shaderop_getrows_f16.h
-          shaderop_getrows_q4_0.h
-          shaderop_getrows_q4_1.h
-          shaderop_getrows_q6_k.h
-          shaderop_rope_f16.h
-          shaderop_rope_f32.h
-          shaderop_cpy_f16_f16.h
-          shaderop_cpy_f16_f32.h
-          shaderop_cpy_f32_f16.h
-          shaderop_cpy_f32_f32.h
-        )
+    # Create a custom target for our generated shaders
+    add_custom_target(generated_shaders DEPENDS
+      shaderop_scale.h
+      shaderop_scale_8.h
+      shaderop_add.h
+      shaderop_addrow.h
+      shaderop_mul.h
+      shaderop_silu.h
+      shaderop_relu.h
+      shaderop_gelu.h
+      shaderop_softmax.h
+      shaderop_norm.h
+      shaderop_rmsnorm.h
+      shaderop_diagmask.h
+      shaderop_mul_mat_mat_f32.h
+      shaderop_mul_mat_f16.h
+      shaderop_mul_mat_q8_0.h
+      shaderop_mul_mat_q4_0.h
+      shaderop_mul_mat_q4_1.h
+      shaderop_mul_mat_q6_k.h
+      shaderop_getrows_f16.h
+      shaderop_getrows_q4_0.h
+      shaderop_getrows_q4_1.h
+      shaderop_getrows_q6_k.h
+      shaderop_rope_f16.h
+      shaderop_rope_f32.h
+      shaderop_cpy_f16_f16.h
+      shaderop_cpy_f16_f32.h
+      shaderop_cpy_f32_f16.h
+      shaderop_cpy_f32_f32.h
+    )
 
-        # Create a custom command that depends on the generated_shaders
-        add_custom_command(
-            OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp
-            COMMAND ${CMAKE_COMMAND} -E touch ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp
-            DEPENDS generated_shaders
-            COMMENT "Ensuring shaders are generated before compiling ggml-kompute.cpp"
-        )
+    # Create a custom command that depends on the generated_shaders
+    add_custom_command(
+        OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp
+        COMMAND ${CMAKE_COMMAND} -E touch ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp
+        DEPENDS generated_shaders
+        COMMENT "Ensuring shaders are generated before compiling ggml-kompute.cpp"
+    )
 
-        # Add the stamp to the main sources to ensure dependency tracking
-        set(GGML_SOURCES_KOMPUTE ${LLAMA_DIR}/ggml-kompute.cpp ${LLAMA_DIR}/ggml-kompute.h ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp)
-        add_compile_definitions(GGML_USE_KOMPUTE)
-        set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} kompute)
-        set(LLAMA_EXTRA_INCLUDES ${LLAMA_EXTRA_INCLUDES} ${CMAKE_BINARY_DIR})
-    else()
-        message(WARNING "Kompute not found")
-    endif()
+    # Add the stamp to the main sources to ensure dependency tracking
+    set(GGML_SOURCES_KOMPUTE ${LLAMA_DIR}/ggml-kompute.cpp ${LLAMA_DIR}/ggml-kompute.h ${CMAKE_CURRENT_BINARY_DIR}/ggml-kompute.stamp)
+    add_compile_definitions(GGML_USE_KOMPUTE)
+    set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} kompute)
+    set(LLAMA_EXTRA_INCLUDES ${LLAMA_EXTRA_INCLUDES} ${CMAKE_BINARY_DIR})
 endif()
 
 if (LLAMA_ALL_WARNINGS)

gpt4all-backend/llamamodel.cpp

@@ -432,6 +432,8 @@ std::vector<LLModel::GPUDevice> LLamaModel::availableGPUDevices(size_t memoryReq
         free(vkDevices);
         return devices;
     }
+#else
+    std::cerr << __func__ << ": built without Kompute\n";
 #endif
 
     return {};

gpt4all-bindings/python/setup.py

@@ -61,7 +61,7 @@ copy_prebuilt_C_lib(SRC_CLIB_DIRECTORY,
 
 setup(
     name=package_name,
-    version="2.2.0",
+    version="2.2.1",
     description="Python bindings for GPT4All",
     author="Nomic and the Open Source Community",
     author_email="support@nomic.ai",