feat(linux): safe resource free up

- Linux x86_64 build setup

Author: Daniele Briggi

.github/workflows/main.yml

@@ -22,7 +22,7 @@ jobs:
           - os: LinuxARM64
             arch: arm64
             name: linux-vulkan
-            make: LLAMA="-DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DGGML_VULKAN=ON"
+            make: LLAMA="-DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_VULKAN=ON"
           - os: macos-latest
             name: macos
           - os: windows-latest
@@ -63,13 +63,24 @@ jobs:
             mingw-w64-x86_64-cmake
             make
 
+      - name: linux install dependencies
+        if: matrix.name == 'linux-vulkan'
+        run: |
+          sudo apt-get update -y
+          sudo apt-get install -y build-essential wget git cmake ccache sqlite3 libcurl4-openssl-dev
+
       - name: linux install vulkan
         if: matrix.name == 'linux-vulkan'
         run: |
           wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo apt-key add -
           sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list
           sudo apt-get update -y
-          sudo apt-get install -y build-essential mesa-vulkan-drivers vulkan-sdk libcurl4-openssl-dev
+          sudo apt-get install -y mesa-vulkan-drivers
+          # Vulkan is no longer packed for Ubuntu
+          wget https://sdk.lunarg.com/sdk/download/latest/linux/vulkan-sdk.tar.xz?Human=true -O vulkan-sdk.tar.xz
+          tar -xf vulkan-sdk.tar.xz
+          cd $(ls -d 1.* | head -n1)
+          echo "source $(pwd)/setup-env.sh" >> ~/.bashrc
 
       - name: windows build sqlite-ai
         if: matrix.name == 'windows'

Makefile

@@ -28,7 +28,7 @@ MAKEFLAGS += -j$(CPUS)
 CC = gcc
 CXX = g++
 CFLAGS = -Wall -Wextra -Wno-unused-parameter -I$(SRC_DIR) -I$(LLAMA_DIR)/ggml/include -I$(LLAMA_DIR)/include
-LDFLAGS = -L./$(BUILD_LLAMA)/common -L./$(BUILD_LLAMA)/ggml/src -L./$(BUILD_LLAMA)/src -L./$(BUILD_WHISPER)/src -lcommon -lggml -lggml-base -lggml-cpu -lllama -lwhisper
+LDFLAGS = -L./$(BUILD_LLAMA)/common -L./$(BUILD_LLAMA)/ggml/src -L./$(BUILD_LLAMA)/src -L./$(BUILD_WHISPER)/src -lcommon -lggml -lggml-base -lllama -lwhisper
 LLAMA_OPTIONS = $(LLAMA) -DLLAMA_CURL=OFF -DLLAMA_BUILD_EXAMPLES=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_TOOLS=OFF -DLLAMA_BUILD_SERVER=OFF
 WHISPER_OPTIONS = $(WHISPER) -DWHISPER_BUILD_EXAMPLES=OFF -DWHISPER_BUILD_TESTS=OFF -DWHISPER_BUILD_SERVER=OFF
 
@@ -85,31 +85,31 @@ else ifeq ($(PLATFORM),android)
     CXX = $(CC)++
     TARGET := $(DIST_DIR)/ai.so
     LDFLAGS += -static-libstdc++ -shared
-    LLAMA_OPTIONS += -DCMAKE_TOOLCHAIN_FILE=$(ANDROID_NDK)/build/cmake/android.toolchain.cmake -DANDROID_ABI=$(if $(filter aarch64,$(ARCH)),arm64-v8a,$(ARCH)) -DANDROID_PLATFORM=android-26 -DCMAKE_C_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DCMAKE_CXX_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DGGML_OPENMP=OFF -DGGML_LLAMAFILE=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    WHISPER_OPTIONS += -DCMAKE_TOOLCHAIN_FILE=$(ANDROID_NDK)/build/cmake/android.toolchain.cmake -DANDROID_ABI=$(if $(filter aarch64,$(ARCH)),arm64-v8a,$(ARCH)) -DANDROID_PLATFORM=android-26 -DCMAKE_C_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DCMAKE_CXX_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DGGML_OPENMP=OFF -DGGML_LLAMAFILE=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+    LLAMA_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_TOOLCHAIN_FILE=$(ANDROID_NDK)/build/cmake/android.toolchain.cmake -DANDROID_ABI=$(if $(filter aarch64,$(ARCH)),arm64-v8a,$(ARCH)) -DANDROID_PLATFORM=android-26 -DCMAKE_C_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DCMAKE_CXX_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DGGML_OPENMP=OFF -DGGML_LLAMAFILE=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+    WHISPER_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_TOOLCHAIN_FILE=$(ANDROID_NDK)/build/cmake/android.toolchain.cmake -DANDROID_ABI=$(if $(filter aarch64,$(ARCH)),arm64-v8a,$(ARCH)) -DANDROID_PLATFORM=android-26 -DCMAKE_C_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DCMAKE_CXX_FLAGS="-march=$(if $(filter aarch64,$(ARCH)),armv8.7a,x86-64)" -DGGML_OPENMP=OFF -DGGML_LLAMAFILE=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
     STRIP = $(BIN)/llvm-strip --strip-unneeded $@
 else ifeq ($(PLATFORM),ios)
     TARGET := $(DIST_DIR)/ai.dylib
     SDK := -isysroot $(shell xcrun --sdk iphoneos --show-sdk-path) -miphoneos-version-min=14.0
     LLAMA_LIBS += $(BUILD_LLAMA)/ggml/src/ggml-metal/libggml-metal.a $(BUILD_LLAMA)/ggml/src/ggml-blas/libggml-blas.a
     LDFLAGS += -L./$(BUILD_LLAMA)/ggml/src/ggml-metal -lggml-metal -L./$(BUILD_LLAMA)/ggml/src/ggml-blas -lggml-blas -framework Accelerate -framework Metal -framework Foundation -framework CoreML -dynamiclib $(SDK)
     CFLAGS += -arch arm64 $(SDK)
-    LLAMA_OPTIONS += -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
-    WHISPER_OPTIONS += -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DWHISPER_COREML=ON
+    LLAMA_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
+    WHISPER_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DWHISPER_COREML=ON
     STRIP = strip -x -S $@
 else ifeq ($(PLATFORM),isim)
     TARGET := $(DIST_DIR)/ai.dylib
     SDK := -isysroot $(shell xcrun --sdk iphonesimulator --show-sdk-path) -miphonesimulator-version-min=14.0
     LLAMA_LIBS += $(BUILD_LLAMA)/ggml/src/ggml-metal/libggml-metal.a $(BUILD_LLAMA)/ggml/src/ggml-blas/libggml-blas.a
     LDFLAGS += -arch x86_64 -arch arm64 -L./$(BUILD_LLAMA)/ggml/src/ggml-metal -lggml-metal -L./$(BUILD_LLAMA)/ggml/src/ggml-blas -lggml-blas -framework Accelerate -framework Metal -framework Foundation -framework CoreML -dynamiclib $(SDK)
     CFLAGS += -arch x86_64 -arch arm64 $(SDK)
-    LLAMA_OPTIONS += -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_SYSROOT=iphonesimulator -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64"
-    WHISPER_OPTIONS += -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_SYSROOT=iphonesimulator -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64" -DWHISPER_COREML=ON
+    LLAMA_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_SYSROOT=iphonesimulator -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64"
+    WHISPER_OPTIONS += -DBUILD_SHARED_LIBS=OFF -DCMAKE_SYSTEM_NAME=iOS -DCMAKE_OSX_SYSROOT=iphonesimulator -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 -DCMAKE_OSX_ARCHITECTURES="x86_64;arm64" -DWHISPER_COREML=ON
 else # linux
     TARGET := $(DIST_DIR)/ai.so
-    LDFLAGS += -shared
-    LLAMA_OPTIONS += -DGGML_OPENMP=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
-    WHISPER_OPTIONS += -DGGML_OPENMP=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+    LDFLAGS += -shared -L./$(BUILD_LLAMA)/bin -Wl,-rpath,./$(BUILD_LLAMA)/bin -Wl,-rpath,./$(BUILD_LLAMA)/common -Wl,-rpath,./$(BUILD_LLAMA)/ggml/src -Wl,-rpath,./$(BUILD_LLAMA)/src -Wl,-rpath,./$(BUILD_WHISPER)/src
+    LLAMA_OPTIONS += -DBUILD_SHARED_LIBS=ON -DGGML_OPENMP=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
+    WHISPER_OPTIONS += -DBUILD_SHARED_LIBS=ON -DGGML_OPENMP=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
     STRIP = strip --strip-unneeded $@
 endif
 
@@ -147,12 +147,12 @@ test: $(TARGET)
 
 # Build submodules
 build/llama.cpp.stamp:
-	cmake -B $(BUILD_LLAMA) -DBUILD_SHARED_LIBS=OFF $(LLAMA_OPTIONS) $(LLAMA_DIR)
+	cmake -B $(BUILD_LLAMA) $(LLAMA_OPTIONS) $(LLAMA_DIR)
 	cmake --build $(BUILD_LLAMA) --config Release -- -j$(CPUS)
 	touch $@
 
 build/whisper.cpp.stamp:
-	cmake -B $(BUILD_WHISPER) -DBUILD_SHARED_LIBS=OFF $(WHISPER_OPTIONS) $(WHISPER_DIR)
+	cmake -B $(BUILD_WHISPER) $(WHISPER_OPTIONS) $(WHISPER_DIR)
 	cmake --build $(BUILD_WHISPER) --config Release -- -j$(CPUS)
 	touch $@
 

src/sqlite-ai.c

@@ -246,6 +246,7 @@ static bool ai_options_callback (sqlite3_context *context, void *xdata, const ch
 void *ai_sqlite_context_create (sqlite3 *db) {
     ai_context *ai = (ai_context *)sqlite3_malloc(sizeof(ai_context));
     if (ai) {
+        memset(ai, 0, sizeof(ai_context));
         ai_options_init(&ai->options);
         ai->db = db;
     }
@@ -258,7 +259,9 @@ void ai_sqlite_context_free (void *ctx) {
     ai_context *ai = (ai_context *)ctx;
     if (ai->model) llama_model_free(ai->model);
     if (ai->ctx) llama_free(ai->ctx);
+    if (ai->sampler) llama_sampler_free(ai->sampler);
     if (ai->token_buffer) sqlite3_free(ai->token_buffer);
+    sqlite3_free(ai);
 }
 
 void ai_logger (enum ggml_log_level level, const char *text, void *user_data) {