llama-cpp-capacitor 0.1.0 → 0.1.2

package/README.md

@@ -121,6 +121,21 @@ cd android
 - **iOS**: `ios/build/LlamaCpp.framework/`
 - **Android**: `android/src/main/jniLibs/{arch}/libllama-cpp-{arch}.so`
 
+### **Updating the native source (e.g. for vision model support)**
+
+The native `cpp/` layer is based on [llama.cpp](https://github.com/ggerganov/llama.cpp). To pull in a newer upstream version (e.g. for vision model support) **without overwriting** the Capacitor adapter code, use the **bootstrap script** (included in this repo):
+
+```bash
+./scripts/bootstrap.sh [branch-or-tag-or-commit]
+# Example: ./scripts/bootstrap.sh master
+```
+
+This syncs upstream into `cpp/` and keeps project-specific files (`cap-*.cpp/h`, `tools/mtmd/`, etc.) intact. After running it, reconcile any API changes in the adapter code, then rebuild with `npm run build:native` or `./build-native.sh`. See [cpp/README.md](cpp/README.md) and [docs/IOS_IMPLEMENTATION_GUIDE.md](docs/IOS_IMPLEMENTATION_GUIDE.md).
+
+### **iOS implementation (step-by-step)**
+
+For a **step-by-step guide** on how methods are implemented on the iOS side (Swift bridge → native framework, adding/updating C symbols, and updating the native layer for vision), see **[docs/IOS_IMPLEMENTATION_GUIDE.md](docs/IOS_IMPLEMENTATION_GUIDE.md)**.
+
 ### iOS Setup
 
 1. Install the plugin:
@@ -696,6 +711,17 @@ const logListener = addNativeLogListener((level, text) => {
 });
 ```
 
+## 📦 Publishing
+
+To publish the package to npm:
+
+1. **Build** (runs automatically on `npm publish` via `prepublishOnly`): `npm run build` — produces `dist/` (plugin bundles, ESM, docs).
+2. **Optional — include native libs in the tarball**: `npm run build:all` (requires macOS/NDK) — builds iOS framework and Android `.so` into `ios/Frameworks` and `android/src/main/jniLibs`.
+3. **Verify pack**: `npm run pack` (JS only) or `npm run pack:full` (JS + native) — lists files that would be published.
+4. **Publish**: `npm publish`.
+
+See [NPM_PUBLISH_GUIDE.md](NPM_PUBLISH_GUIDE.md) for 2FA/token setup and troubleshooting.
+
 ## 🤝 Contributing
 
 We welcome contributions! Please see our [Contributing Guide](CONTRIBUTING.md) for details.

package/build-native.sh

@@ -0,0 +1,229 @@
+#!/bin/bash
+
+# Build script for llama-cpp Capacitor plugin
+# This script compiles the native llama.cpp library for iOS and Android
+
+set -e
+
+# Colors for output
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+BLUE='\033[0;34m'
+NC='\033[0m' # No Color
+
+# Function to print colored output
+print_status() {
+    echo -e "${BLUE}[INFO]${NC} $1"
+}
+
+print_success() {
+    echo -e "${GREEN}[SUCCESS]${NC} $1"
+}
+
+print_warning() {
+    echo -e "${YELLOW}[WARNING]${NC} $1"
+}
+
+print_error() {
+    echo -e "${RED}[ERROR]${NC} $1"
+}
+
+# Check if we're on macOS for iOS builds
+check_macos() {
+    if [[ "$OSTYPE" != "darwin"* ]]; then
+        print_warning "iOS builds require macOS. Skipping iOS build."
+        return 1
+    fi
+    return 0
+}
+
+# Check if Android SDK is available
+check_android_sdk() {
+    if [ -z "$ANDROID_HOME" ] && [ -z "$ANDROID_SDK_ROOT" ]; then
+        print_warning "Android SDK not found. Please set ANDROID_HOME or ANDROID_SDK_ROOT."
+        print_warning "Skipping Android build."
+        return 1
+    fi
+    return 0
+}
+
+# Detect NDK path: ndk/<version>/build/cmake/android.toolchain.cmake
+# Uses latest versioned NDK under $ANDROID_SDK/ndk/ (e.g. ndk/29.0.13113456).
+detect_ndk() {
+    local sdk="${ANDROID_HOME:-$ANDROID_SDK_ROOT}"
+    local ndk_root="$sdk/ndk"
+    if [ ! -d "$ndk_root" ]; then
+        echo ""
+        return 1
+    fi
+    local latest=""
+    local latest_ver=0
+    for v in "$ndk_root"/*; do
+        [ -d "$v" ] || continue
+        local base=$(basename "$v")
+        if [[ "$base" =~ ^[0-9] ]]; then
+            local ver=$(echo "$base" | sed 's/[^0-9]//g' | head -c 10)
+            ver=${ver:-0}
+            if [ "$ver" -gt "$latest_ver" ] 2>/dev/null; then
+                latest_ver=$ver
+                latest=$v
+            fi
+        fi
+    done
+    if [ -n "$latest" ] && [ -f "$latest/build/cmake/android.toolchain.cmake" ]; then
+        echo "$latest"
+        return 0
+    fi
+    echo ""
+    return 1
+}
+
+# Build iOS library
+build_ios() {
+    print_status "Building iOS library..."
+    
+    if ! check_macos; then
+        return 1
+    fi
+    
+    # Always start from a clean iOS build directory to avoid stale Xcode settings
+    rm -rf ios/build
+    mkdir -p ios/build
+    cd ios/build
+    
+    # Configure with CMake
+    # IMPORTANT: build iOS framework as **ARM64-only**.
+    # Including x86_64 here makes CMake/Xcode try to link an x86_64 slice,
+    # but we only compile ARM-specific kernels (arch/arm), which leads to
+    # undefined symbols like lm_ggml_gemm_* for x86_64.
+    cmake .. \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DCMAKE_OSX_ARCHITECTURES="arm64" \
+        -DCMAKE_OSX_DEPLOYMENT_TARGET=13.0 \
+        -DCMAKE_XCODE_ATTRIBUTE_ENABLE_BITCODE=NO
+    
+    # Build
+    cmake --build . --config Release
+    
+    # CMake builds the framework directly (FRAMEWORK TRUE in CMakeLists.txt)
+    # Verify the framework was created
+    if [ -d "llama-cpp.framework" ]; then
+        print_success "iOS framework built successfully at: $(pwd)/llama-cpp.framework"
+        
+        # Strip debug symbols to reduce app store size (~0.5–1 MB)
+        BINARY="llama-cpp.framework/Versions/A/llama-cpp"
+        if [ -f "$BINARY" ]; then
+            if xcrun strip -x -S "$BINARY" 2>/dev/null; then
+                print_status "Stripped debug symbols from iOS framework"
+            fi
+        fi
+        
+        # Copy framework to package location for npm publishing
+        mkdir -p ../Frameworks
+        cp -R llama-cpp.framework ../Frameworks/
+        print_success "iOS framework copied to ios/Frameworks/ for npm package"
+    else
+        print_error "iOS framework not found after build"
+        cd ../..
+        return 1
+    fi
+    
+    cd ../..
+}
+
+# Build Android library
+build_android() {
+    print_status "Building Android library..."
+    
+    if ! check_android_sdk; then
+        return 1
+    fi
+    
+    ANDROID_NDK=$(detect_ndk)
+    if [ -z "$ANDROID_NDK" ]; then
+        print_error "Android NDK not found. Install NDK via Android Studio (SDK Manager → NDK)."
+        print_error "Expected: \$ANDROID_HOME/ndk/<version>/build/cmake/android.toolchain.cmake"
+        return 1
+    fi
+    print_status "Using NDK: $ANDROID_NDK"
+    
+    TOOLCHAIN_FILE="$ANDROID_NDK/build/cmake/android.toolchain.cmake"
+    if [ ! -f "$TOOLCHAIN_FILE" ]; then
+        print_error "Toolchain file not found: $TOOLCHAIN_FILE"
+        return 1
+    fi
+    
+    rm -rf android/build
+    mkdir -p android/build
+    cd android/build
+    
+    # Build only arm64-v8a to minimize app store size.
+    # build.gradle uses abiFilters 'arm64-v8a'; armeabi-v7a is not included in the app.
+    # Skipping armeabi-v7a saves ~25–48 MB in package and keeps app size minimal.
+    arch="arm64-v8a"
+    print_status "Building for $arch (only ABI shipped in app)..."
+    rm -rf CMakeCache.txt CMakeFiles Makefile cmake_install.cmake 2>/dev/null || true
+    find . -maxdepth 1 -name '*.so' -delete 2>/dev/null || true
+    
+    cmake ../src/main \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DANDROID_ABI=$arch \
+        -DANDROID_PLATFORM=android-21 \
+        -DCMAKE_TOOLCHAIN_FILE="$TOOLCHAIN_FILE" \
+        -DANDROID_STL=c++_shared
+    
+    cmake --build . --config Release
+    
+    mkdir -p ../src/main/jniLibs/$arch
+    if [ -f "libllama-cpp-arm64.so" ]; then
+        PREBUILT="$ANDROID_NDK/toolchains/llvm/prebuilt"
+        if [ -d "$PREBUILT/darwin-x86_64" ]; then
+            STRIP_TOOL="$PREBUILT/darwin-x86_64/bin/llvm-strip"
+        elif [ -d "$PREBUILT/darwin-aarch64" ]; then
+            STRIP_TOOL="$PREBUILT/darwin-aarch64/bin/llvm-strip"
+        else
+            STRIP_TOOL=$(find "$ANDROID_NDK/toolchains" -name "llvm-strip" -type f 2>/dev/null | head -1)
+        fi
+        if [ -f "$STRIP_TOOL" ]; then
+            "$STRIP_TOOL" --strip-debug libllama-cpp-arm64.so
+            print_status "Stripped debug symbols from library"
+        fi
+        cp "libllama-cpp-arm64.so" "../src/main/jniLibs/$arch/"
+    fi
+    print_success "Built for $arch"
+    
+    print_success "Android library built successfully"
+    cd ../..
+}
+
+# Main build function
+main() {
+    print_status "Starting llama-cpp Capacitor plugin build..."
+    
+    # Check dependencies
+    if ! command -v cmake &> /dev/null; then
+        print_error "CMake is required but not installed"
+        exit 1
+    fi
+    
+    if ! command -v make &> /dev/null; then
+        print_error "Make is required but not installed"
+        exit 1
+    fi
+    
+    # Build iOS
+    if check_macos; then
+        build_ios
+    fi
+    
+    # Build Android
+    if check_android_sdk; then
+        build_android
+    fi
+    
+    print_success "Build completed successfully!"
+}
+
+# Run main function
+main "$@"

package/cpp/README.md

@@ -1,4 +1,15 @@
 # Note
 
-- Only `rn-tts.h`, `rn-tts.cpp`, `rn-mtmd.hpp`, `rn-mtmd.cpp`, `rn-llama.h` and `rn-llama.cpp` are the specific files for this folder, others are sync from [llama.cpp](https://github.com/ggerganov/llama.cpp).
-- We can update the native source by using the [bootstrap](../scripts/bootstrap.sh) script.
+- The **project-specific** files in this folder (never overwritten by bootstrap) are:
+  - `cap-llama.cpp` / `cap-llama.h` – Capacitor bridge (context, model load)
+  - `cap-completion.cpp` / `cap-completion.h` – completion
+  - `cap-tts.cpp` / `cap-tts.h` – TTS
+  - `cap-embedding.cpp` / `cap-embedding.h` – embeddings
+  - `cap-mtmd.hpp` – multimodal/vision
+  - `tools/mtmd/` – multimodal/vision tooling
+- All other sources are synced from [llama.cpp](https://github.com/ggerganov/llama.cpp).
+- **Update native source** (e.g. for a newer llama.cpp or vision support): run from repo root:
+  ```bash
+  ./scripts/bootstrap.sh [branch-or-tag-or-commit]
+  ```
+  See [scripts/bootstrap.sh](../scripts/bootstrap.sh) and [docs/IOS_IMPLEMENTATION_GUIDE.md](../docs/IOS_IMPLEMENTATION_GUIDE.md).

package/ios/CMakeLists-arm64.txt

@@ -0,0 +1,149 @@
+cmake_minimum_required(VERSION 3.16)
+project(llama-cpp VERSION 1.0.0 LANGUAGES CXX C)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+# iOS specific settings for ARM64 devices
+set(CMAKE_OSX_DEPLOYMENT_TARGET 13.0)
+set(CMAKE_XCODE_ATTRIBUTE_ENABLE_BITCODE NO)
+
+# Common llama.cpp definitions
+add_definitions(
+    -DNDEBUG
+    -DO3
+    -DLM_GGML_USE_CPU
+    -DLM_GGML_USE_ACCELERATE
+    -DLM_GGML_USE_METAL
+    -DLM_GGML_METAL_USE_BF16
+)
+
+set(SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../cpp)
+
+# ARM64‑tuned ggml sources
+set(SOURCE_FILES_ARCH
+    ${SOURCE_DIR}/ggml-cpu/arch/arm/quants.c
+    ${SOURCE_DIR}/ggml-cpu/arch/arm/repack.cpp
+)
+
+# Define public headers
+set(PUBLIC_HEADERS
+    ${SOURCE_DIR}/cap-llama.h
+    ${SOURCE_DIR}/cap-completion.h
+    ${SOURCE_DIR}/cap-tts.h
+    ${SOURCE_DIR}/llama.h
+    ${SOURCE_DIR}/ggml.h
+)
+
+# Create library target
+add_library(llama-cpp SHARED
+    ${SOURCE_DIR}/ggml.c
+    ${SOURCE_DIR}/ggml-alloc.c
+    ${SOURCE_DIR}/ggml-backend.cpp
+    ${SOURCE_DIR}/ggml-backend-reg.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/amx.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/mmq.cpp
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.c
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.cpp
+    ${SOURCE_DIR}/ggml-cpu/quants.c
+    ${SOURCE_DIR}/ggml-cpu/traits.cpp
+    ${SOURCE_DIR}/ggml-cpu/repack.cpp
+    ${SOURCE_DIR}/ggml-cpu/unary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/binary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/vec.cpp
+    ${SOURCE_DIR}/ggml-cpu/ops.cpp
+    ${SOURCE_DIR}/ggml-metal.m
+    ${SOURCE_DIR}/ggml-opt.cpp
+    ${SOURCE_DIR}/ggml-threading.cpp
+    ${SOURCE_DIR}/ggml-quants.c
+    ${SOURCE_DIR}/gguf.cpp
+    ${SOURCE_DIR}/log.cpp
+    ${SOURCE_DIR}/llama-impl.cpp
+    ${SOURCE_DIR}/llama-grammar.cpp
+    ${SOURCE_DIR}/llama-sampling.cpp
+    ${SOURCE_DIR}/llama-vocab.cpp
+    ${SOURCE_DIR}/llama-adapter.cpp
+    ${SOURCE_DIR}/llama-chat.cpp
+    ${SOURCE_DIR}/llama-context.cpp
+    ${SOURCE_DIR}/llama-arch.cpp
+    ${SOURCE_DIR}/llama-batch.cpp
+    ${SOURCE_DIR}/llama-cparams.cpp
+    ${SOURCE_DIR}/llama-hparams.cpp
+    ${SOURCE_DIR}/llama.cpp
+    ${SOURCE_DIR}/llama-model.cpp
+    ${SOURCE_DIR}/llama-model-loader.cpp
+    ${SOURCE_DIR}/llama-model-saver.cpp
+    ${SOURCE_DIR}/llama-mmap.cpp
+    ${SOURCE_DIR}/llama-kv-cache.cpp
+    ${SOURCE_DIR}/llama-kv-cache-iswa.cpp
+    ${SOURCE_DIR}/llama-memory-hybrid.cpp
+    ${SOURCE_DIR}/llama-memory-recurrent.cpp
+    ${SOURCE_DIR}/llama-memory.cpp
+    ${SOURCE_DIR}/llama-io.cpp
+    ${SOURCE_DIR}/llama-graph.cpp
+    ${SOURCE_DIR}/sampling.cpp
+    ${SOURCE_DIR}/unicode-data.cpp
+    ${SOURCE_DIR}/unicode.cpp
+    ${SOURCE_DIR}/common.cpp
+    ${SOURCE_DIR}/chat.cpp
+    # Additional sources required for JSON parsing, chat parser, and mtmd tools
+    ${SOURCE_DIR}/chat-parser.cpp
+    ${SOURCE_DIR}/regex-partial.cpp
+    ${SOURCE_DIR}/json-partial.cpp
+    ${SOURCE_DIR}/json-schema-to-grammar.cpp
+    ${SOURCE_DIR}/anyascii.c
+    ${SOURCE_DIR}/tools/mtmd/mtmd.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-helper.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-audio.cpp
+    ${SOURCE_DIR}/tools/mtmd/clip.cpp
+    ${SOURCE_DIR}/cap-llama.cpp
+    ${SOURCE_DIR}/cap-completion.cpp
+    ${SOURCE_DIR}/cap-tts.cpp
+    ${SOURCE_FILES_ARCH}
+)
+
+# Set target properties
+set_target_properties(llama-cpp PROPERTIES
+    FRAMEWORK TRUE
+    FRAMEWORK_VERSION A
+    MACOSX_FRAMEWORK_IDENTIFIER com.arusatech.llama-cpp
+    MACOSX_FRAMEWORK_BUNDLE_VERSION 1.0.0
+    MACOSX_FRAMEWORK_SHORT_VERSION_STRING 1.0.0
+    XCODE_ATTRIBUTE_CODE_SIGN_IDENTITY "iPhone Developer"
+    XCODE_ATTRIBUTE_DEVELOPMENT_TEAM ""
+    XCODE_ATTRIBUTE_ONLY_ACTIVE_ARCH NO
+    XCODE_ATTRIBUTE_ENABLE_BITCODE NO
+)
+
+# Include directories
+target_include_directories(llama-cpp PRIVATE
+    ${SOURCE_DIR}
+    ${SOURCE_DIR}/ggml-cpu
+    ${SOURCE_DIR}/tools/mtmd
+)
+
+# ARM64: rely on Clang's default tuning; no x86‑specific -march flags here
+
+# Link frameworks via linker flags
+target_link_options(llama-cpp PRIVATE
+    "-Wl,-framework,Accelerate"
+    "-Wl,-framework,Metal"
+    "-Wl,-framework,MetalKit"
+    "-Wl,-framework,Foundation"
+    "-Wl,-framework,CoreGraphics"
+)
+
+# Public headers
+set_target_properties(llama-cpp PROPERTIES
+    PUBLIC_HEADER "${PUBLIC_HEADERS}"
+)
+
+# Install rules
+install(TARGETS llama-cpp
+    FRAMEWORK DESTINATION .
+)
+
+message(STATUS "Building llama-cpp for ARM64 (devices)")
+message(STATUS "Source directory: ${SOURCE_DIR}")
+message(STATUS "Architecture: arm64")
+

package/ios/CMakeLists-x86_64.txt

@@ -0,0 +1,176 @@
+cmake_minimum_required(VERSION 3.16)
+project(llama-cpp VERSION 1.0.0 LANGUAGES CXX C)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+# iOS specific settings for x86_64 emulator
+set(CMAKE_OSX_DEPLOYMENT_TARGET 13.0)
+set(CMAKE_XCODE_ATTRIBUTE_ENABLE_BITCODE NO)
+
+# Dependencies and compile options optimized for x86_64
+add_definitions(
+    -DNDEBUG
+    -DO3
+    -DLM_GGML_USE_CPU
+    -DLM_GGML_USE_ACCELERATE
+    -DLM_GGML_USE_METAL
+    -DLM_GGML_METAL_USE_BF16
+)
+
+# X86_64 specific optimizations for emulator
+add_definitions(-DLM_GGML_CPU_GENERIC)
+add_definitions(-DLM_GGML_USE_AVX2)
+add_definitions(-DLM_GGML_USE_AVX)
+add_definitions(-DLM_GGML_USE_SSE3)
+add_definitions(-DLM_GGML_USE_SSE)
+add_definitions(-DLM_GGML_USE_F16C)
+add_definitions(-DLM_GGML_USE_FMA)
+
+set(SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../cpp)
+
+# Use x86_64 optimized source files for emulator
+set(SOURCE_FILES_ARCH
+    ${SOURCE_DIR}/ggml-cpu/arch/x86/quants.c
+    ${SOURCE_DIR}/ggml-cpu/arch/x86/repack.cpp
+)
+
+# Define public headers
+set(PUBLIC_HEADERS
+    ${SOURCE_DIR}/cap-llama.h
+    ${SOURCE_DIR}/cap-completion.h
+    ${SOURCE_DIR}/cap-tts.h
+    ${SOURCE_DIR}/llama.h
+    ${SOURCE_DIR}/ggml.h
+)
+
+# Create library target
+add_library(llama-cpp SHARED
+    ${SOURCE_DIR}/ggml.c
+    ${SOURCE_DIR}/ggml-alloc.c
+    ${SOURCE_DIR}/ggml-backend.cpp
+    ${SOURCE_DIR}/ggml-backend-reg.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/amx.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/mmq.cpp
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.c
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.cpp
+    ${SOURCE_DIR}/ggml-cpu/quants.c
+    ${SOURCE_DIR}/ggml-cpu/traits.cpp
+    ${SOURCE_DIR}/ggml-cpu/repack.cpp
+    ${SOURCE_DIR}/ggml-cpu/unary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/binary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/vec.cpp
+    ${SOURCE_DIR}/ggml-cpu/ops.cpp
+    ${SOURCE_DIR}/ggml-metal.m
+    ${SOURCE_DIR}/ggml-opt.cpp
+    ${SOURCE_DIR}/ggml-threading.cpp
+    ${SOURCE_DIR}/ggml-quants.c
+    ${SOURCE_DIR}/gguf.cpp
+    ${SOURCE_DIR}/log.cpp
+    ${SOURCE_DIR}/llama-impl.cpp
+    ${SOURCE_DIR}/llama-grammar.cpp
+    ${SOURCE_DIR}/llama-sampling.cpp
+    ${SOURCE_DIR}/llama-vocab.cpp
+    ${SOURCE_DIR}/llama-adapter.cpp
+    ${SOURCE_DIR}/llama-chat.cpp
+    ${SOURCE_DIR}/llama-context.cpp
+    ${SOURCE_DIR}/llama-arch.cpp
+    ${SOURCE_DIR}/llama-batch.cpp
+    ${SOURCE_DIR}/llama-cparams.cpp
+    ${SOURCE_DIR}/llama-hparams.cpp
+    ${SOURCE_DIR}/llama.cpp
+    ${SOURCE_DIR}/llama-model.cpp
+    ${SOURCE_DIR}/llama-model-loader.cpp
+    ${SOURCE_DIR}/llama-model-saver.cpp
+    ${SOURCE_DIR}/llama-mmap.cpp
+    ${SOURCE_DIR}/llama-kv-cache.cpp
+    ${SOURCE_DIR}/llama-kv-cache-iswa.cpp
+    ${SOURCE_DIR}/llama-memory-hybrid.cpp
+    ${SOURCE_DIR}/llama-memory-recurrent.cpp
+    ${SOURCE_DIR}/llama-memory.cpp
+    ${SOURCE_DIR}/llama-io.cpp
+    ${SOURCE_DIR}/llama-graph.cpp
+    ${SOURCE_DIR}/sampling.cpp
+    ${SOURCE_DIR}/unicode-data.cpp
+    ${SOURCE_DIR}/unicode.cpp
+    ${SOURCE_DIR}/common.cpp
+    ${SOURCE_DIR}/chat.cpp
+    # Additional sources required for JSON parsing, chat parser, and mtmd tools
+    ${SOURCE_DIR}/chat-parser.cpp
+    ${SOURCE_DIR}/regex-partial.cpp
+    ${SOURCE_DIR}/json-partial.cpp
+    ${SOURCE_DIR}/json-schema-to-grammar.cpp
+    ${SOURCE_DIR}/anyascii.c
+    ${SOURCE_DIR}/tools/mtmd/mtmd.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-helper.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-audio.cpp
+    ${SOURCE_DIR}/tools/mtmd/clip.cpp
+    ${SOURCE_DIR}/minja/minja.hpp
+    ${SOURCE_DIR}/minja/chat-template.hpp
+    ${SOURCE_DIR}/nlohmann/json.hpp
+    ${SOURCE_DIR}/nlohmann/json_fwd.hpp
+    ${SOURCE_DIR}/cap-llama.cpp
+    ${SOURCE_DIR}/cap-completion.cpp
+    ${SOURCE_DIR}/cap-tts.cpp
+    ${SOURCE_FILES_ARCH}
+)
+
+# Set target properties
+set_target_properties(llama-cpp PROPERTIES
+    FRAMEWORK TRUE
+    FRAMEWORK_VERSION A
+    MACOSX_FRAMEWORK_IDENTIFIER com.arusatech.llama-cpp
+    MACOSX_FRAMEWORK_BUNDLE_VERSION 1.0.0
+    MACOSX_FRAMEWORK_SHORT_VERSION_STRING 1.0.0
+    XCODE_ATTRIBUTE_CODE_SIGN_IDENTITY "iPhone Developer"
+    XCODE_ATTRIBUTE_DEVELOPMENT_TEAM ""
+    XCODE_ATTRIBUTE_ONLY_ACTIVE_ARCH NO
+    XCODE_ATTRIBUTE_ENABLE_BITCODE NO
+)
+
+# Set include directories
+target_include_directories(llama-cpp PRIVATE
+    ${SOURCE_DIR}
+    ${SOURCE_DIR}/ggml-cpu
+    ${SOURCE_DIR}/tools/mtmd
+)
+
+# Set compile options for x86_64 optimization
+target_compile_options(llama-cpp PRIVATE
+    -march=x86-64
+    -mtune=generic
+    -mavx2
+    -mavx
+    -msse3
+    -msse
+    -mfma
+    -mf16c
+)
+
+# Set link options
+# NOTE:
+# Use -Wl,-framework,<Name> so that the compiler driver passes the correct
+# framework flags through to the linker as a single option.
+target_link_options(llama-cpp PRIVATE
+    "-Wl,-framework,Accelerate"
+    "-Wl,-framework,Metal"
+    "-Wl,-framework,MetalKit"
+    "-Wl,-framework,Foundation"
+    "-Wl,-framework,CoreGraphics"
+)
+
+# Set public headers
+set_target_properties(llama-cpp PROPERTIES
+    PUBLIC_HEADER "${PUBLIC_HEADERS}"
+)
+
+# Install rules
+install(TARGETS llama-cpp
+    FRAMEWORK DESTINATION .
+)
+
+# Print build information
+message(STATUS "Building llama-cpp for x86_64 (emulator)")
+message(STATUS "Source directory: ${SOURCE_DIR}")
+message(STATUS "Architecture: x86_64")
+message(STATUS "Optimizations: AVX2, AVX, SSE3, SSE, FMA, F16C")

package/ios/CMakeLists.txt

@@ -0,0 +1,149 @@
+cmake_minimum_required(VERSION 3.16)
+project(llama-cpp VERSION 1.0.0 LANGUAGES CXX C)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+# iOS specific settings for ARM64 devices
+set(CMAKE_OSX_DEPLOYMENT_TARGET 13.0)
+set(CMAKE_XCODE_ATTRIBUTE_ENABLE_BITCODE NO)
+
+# Common llama.cpp definitions
+add_definitions(
+    -DNDEBUG
+    -DO3
+    -DLM_GGML_USE_CPU
+    -DLM_GGML_USE_ACCELERATE
+    -DLM_GGML_USE_METAL
+    -DLM_GGML_METAL_USE_BF16
+)
+
+set(SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/../cpp)
+
+# ARM64‑tuned ggml sources
+set(SOURCE_FILES_ARCH
+    ${SOURCE_DIR}/ggml-cpu/arch/arm/quants.c
+    ${SOURCE_DIR}/ggml-cpu/arch/arm/repack.cpp
+)
+
+# Define public headers
+set(PUBLIC_HEADERS
+    ${SOURCE_DIR}/cap-llama.h
+    ${SOURCE_DIR}/cap-completion.h
+    ${SOURCE_DIR}/cap-tts.h
+    ${SOURCE_DIR}/llama.h
+    ${SOURCE_DIR}/ggml.h
+)
+
+# Create library target
+add_library(llama-cpp SHARED
+    ${SOURCE_DIR}/ggml.c
+    ${SOURCE_DIR}/ggml-alloc.c
+    ${SOURCE_DIR}/ggml-backend.cpp
+    ${SOURCE_DIR}/ggml-backend-reg.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/amx.cpp
+    ${SOURCE_DIR}/ggml-cpu/amx/mmq.cpp
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.c
+    ${SOURCE_DIR}/ggml-cpu/ggml-cpu.cpp
+    ${SOURCE_DIR}/ggml-cpu/quants.c
+    ${SOURCE_DIR}/ggml-cpu/traits.cpp
+    ${SOURCE_DIR}/ggml-cpu/repack.cpp
+    ${SOURCE_DIR}/ggml-cpu/unary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/binary-ops.cpp
+    ${SOURCE_DIR}/ggml-cpu/vec.cpp
+    ${SOURCE_DIR}/ggml-cpu/ops.cpp
+    ${SOURCE_DIR}/ggml-metal.m
+    ${SOURCE_DIR}/ggml-opt.cpp
+    ${SOURCE_DIR}/ggml-threading.cpp
+    ${SOURCE_DIR}/ggml-quants.c
+    ${SOURCE_DIR}/gguf.cpp
+    ${SOURCE_DIR}/log.cpp
+    ${SOURCE_DIR}/llama-impl.cpp
+    ${SOURCE_DIR}/llama-grammar.cpp
+    ${SOURCE_DIR}/llama-sampling.cpp
+    ${SOURCE_DIR}/llama-vocab.cpp
+    ${SOURCE_DIR}/llama-adapter.cpp
+    ${SOURCE_DIR}/llama-chat.cpp
+    ${SOURCE_DIR}/llama-context.cpp
+    ${SOURCE_DIR}/llama-arch.cpp
+    ${SOURCE_DIR}/llama-batch.cpp
+    ${SOURCE_DIR}/llama-cparams.cpp
+    ${SOURCE_DIR}/llama-hparams.cpp
+    ${SOURCE_DIR}/llama.cpp
+    ${SOURCE_DIR}/llama-model.cpp
+    ${SOURCE_DIR}/llama-model-loader.cpp
+    ${SOURCE_DIR}/llama-model-saver.cpp
+    ${SOURCE_DIR}/llama-mmap.cpp
+    ${SOURCE_DIR}/llama-kv-cache.cpp
+    ${SOURCE_DIR}/llama-kv-cache-iswa.cpp
+    ${SOURCE_DIR}/llama-memory-hybrid.cpp
+    ${SOURCE_DIR}/llama-memory-recurrent.cpp
+    ${SOURCE_DIR}/llama-memory.cpp
+    ${SOURCE_DIR}/llama-io.cpp
+    ${SOURCE_DIR}/llama-graph.cpp
+    ${SOURCE_DIR}/sampling.cpp
+    ${SOURCE_DIR}/unicode-data.cpp
+    ${SOURCE_DIR}/unicode.cpp
+    ${SOURCE_DIR}/common.cpp
+    ${SOURCE_DIR}/chat.cpp
+    # Additional sources required for JSON parsing, chat parser, and mtmd tools
+    ${SOURCE_DIR}/chat-parser.cpp
+    ${SOURCE_DIR}/regex-partial.cpp
+    ${SOURCE_DIR}/json-partial.cpp
+    ${SOURCE_DIR}/json-schema-to-grammar.cpp
+    ${SOURCE_DIR}/anyascii.c
+    ${SOURCE_DIR}/tools/mtmd/mtmd.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-helper.cpp
+    ${SOURCE_DIR}/tools/mtmd/mtmd-audio.cpp
+    ${SOURCE_DIR}/tools/mtmd/clip.cpp
+    ${SOURCE_DIR}/cap-llama.cpp
+    ${SOURCE_DIR}/cap-completion.cpp
+    ${SOURCE_DIR}/cap-tts.cpp
+    ${SOURCE_FILES_ARCH}
+)
+
+# Set target properties
+set_target_properties(llama-cpp PROPERTIES
+    FRAMEWORK TRUE
+    FRAMEWORK_VERSION A
+    MACOSX_FRAMEWORK_IDENTIFIER com.arusatech.llama-cpp
+    MACOSX_FRAMEWORK_BUNDLE_VERSION 1.0.0
+    MACOSX_FRAMEWORK_SHORT_VERSION_STRING 1.0.0
+    XCODE_ATTRIBUTE_CODE_SIGN_IDENTITY "iPhone Developer"
+    XCODE_ATTRIBUTE_DEVELOPMENT_TEAM ""
+    XCODE_ATTRIBUTE_ONLY_ACTIVE_ARCH NO
+    XCODE_ATTRIBUTE_ENABLE_BITCODE NO
+)
+
+# Include directories
+target_include_directories(llama-cpp PRIVATE
+    ${SOURCE_DIR}
+    ${SOURCE_DIR}/ggml-cpu
+    ${SOURCE_DIR}/tools/mtmd
+)
+
+# ARM64: rely on Clang's default tuning; no x86‑specific -march flags here
+
+# Link frameworks via linker flags
+target_link_options(llama-cpp PRIVATE
+    "-Wl,-framework,Accelerate"
+    "-Wl,-framework,Metal"
+    "-Wl,-framework,MetalKit"
+    "-Wl,-framework,Foundation"
+    "-Wl,-framework,CoreGraphics"
+)
+
+# Public headers
+set_target_properties(llama-cpp PROPERTIES
+    PUBLIC_HEADER "${PUBLIC_HEADERS}"
+)
+
+# Install rules
+install(TARGETS llama-cpp
+    FRAMEWORK DESTINATION .
+)
+
+message(STATUS "Building llama-cpp for ARM64 (devices)")
+message(STATUS "Source directory: ${SOURCE_DIR}")
+message(STATUS "Architecture: arm64")
+

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "llama-cpp-capacitor",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "A native Capacitor plugin that embeds llama.cpp directly into mobile apps, enabling offline AI inference with chat-first API design. Complete iOS and Android support: text generation, chat, multimodal, TTS, LoRA, embeddings, and more.",
   "main": "dist/plugin.cjs.js",
   "type": "module",
@@ -10,11 +10,14 @@
   "files": [
     "android/src/main/",
     "android/build.gradle",
+    "build-native.sh",
     "cpp/",
     "dist/",
+    "ios/CMakeLists.txt",
+    "ios/CMakeLists-arm64.txt",
+    "ios/CMakeLists-x86_64.txt",
     "ios/Sources",
     "ios/Frameworks",
-    "ios/Tests",
     "Package.swift",
     "LlamaCpp.podspec",
     "types/"
@@ -65,8 +68,11 @@
     "docgen": "npx docgen --api LlamaCppPlugin --output-readme README.md --output-json dist/docs.json",
     "build": "npm run clean && npm run docgen && tsc && rollup -c rollup.config.mjs",
     "build:native": "./build-native.sh",
+    "build:all": "npm run build && npm run build:native",
     "build:ios": "cd ios && cmake -B build -S . && cmake --build build --config Release",
     "build:android": "cd android && gradlew.bat assembleRelease",
+    "pack": "npm run build && npm pack --dry-run",
+    "pack:full": "npm run build:all && npm pack --dry-run",
     "test": "jest",
     "test:integration": "./scripts/test-integration.sh",
     "test:integration:jest": "jest --config test/jest.integration.config.js",

package/ios/Tests/LlamaCppPluginTests/LlamaCppPluginTests.swift

@@ -1,15 +0,0 @@
-import XCTest
-@testable import LlamaCppPlugin
-
-class LlamaCppTests: XCTestCase {
-    func testEcho() {
-        // This is an example of a functional test case for a plugin.
-        // Use XCTAssert and related functions to verify your tests produce the correct results.
-
-        let implementation = LlamaCpp()
-        let value = "Hello, World!"
-        let result = implementation.echo(value)
-
-        XCTAssertEqual(value, result)
-    }
-}