The swiftui DEMO's build-xcframework.sh doesn't work as expected

[tokyo4]

hello

follow the step but it doesn't wok as expected

I already run "sudo xcode-select -switch /Applications/Xcode.app/Contents/Developer" command

but still got those error messages

thx

my device is Mac Mini M4

[jenhseb]

Am getting this too

[jenhseb]

@tokyo4 found this gem in the adjacent repo. Was able to resolve this error in my case by following this tip here
ggml-org/llama.cpp#11996 (comment)

[danbev]

There is now a built xcframework attached to the latest release which can be used directly if you don't want to build yourself:
https://github.com/ggml-org/whisper.cpp/releases

[tokyo4]

@tokyo4 found this gem in the adjacent repo. Was able to resolve this error in my case by following this tip here ggml-org/llama.cpp#11996 (comment)

thx for your replay,but I don't know how .
is that i need to remove the llama.cpp in the example/talk-llama/ and run ./build-xcframework.sh?

[danbev]

By the title of this discussion I thought you were using the whisper.swiftui example.

I have not tried talk-llama in combination with whisper.xcframework, but if you have created an xcode project you can download and unzip the the xcframework from the release, and then add the framework to your project in Frameworks, Libraries, and Embedded Content. Perhaps try this with the whisper.swiftui example first just to verify that it works and how the framework is added (there are instructions in the examples README.md).

[jenhseb]

@tokyo4 found this gem in the adjacent repo. Was able to resolve this error in my case by following this tip here ggml-org/llama.cpp#11996 (comment)

thx for your replay,but I don't know how . is that i need to remove the llama.cpp in the example/talk-llama/ and run ./build-xcframework.sh?

The linked comment resolved a similar issue someone experienced with the llama cpp build by cloning the repo outside of an iCloud folder and building it there. I did the same with the whisper cpp repo and was able to resolve the whisper error this way as well

[tokyo4]

I have not tried talk-llama in combination with whisper.xcframework, but if you have created an xcode project you can download and unzip the the xcframework from the release, and then add the framework to your project in Frameworks, Libraries, and Embedded Content. Perhaps try this with the whisper.swiftui example first just to verify that it works and how the framework is added (there are instructions in the examples README.md).

thx，follow your link, I downloaded the project and xcframework in my local disk, and run "./build-xcframework.sh" once on my terminal, and then xcode build the demo app successed in my iphone 16 promax

[StefanBristol]

I have built an app using Xcode that calls on a couple of packages and now includes llama.xcframework (created from the updated .sh script that uses cmake, xcrun and xcbuild), instead of the earlier packaged version. I've included the script below - I too had issues getting the provided script to work.
THEN
After creating the llama.xcframework folder - copy it into your project folder in Finder and make sure llama.xcframework appears in Frameworks, Libraries and Embedded Content in your TARGET (app) Build Settings.
THEN
Find the created default.metallib file and drag it from Finder into your Xcode Navigation view, under Resources. Make sure it appears (or add) under Copy Bundle Resources in you TARGET Build Phases.

My issue...
The app builds and runs as expected in Xcode simulator and on an iPad device. However, when an Archive is created, neither Validate App or Distribute App will work. The following error returned is...

Failed Platform Validation "MyAppName" requires the "iPhoneOS" platform to be installed.

Has anyone else had this issue? I've read this thread with no resolution.
The only real difference that I've spotted is that I do not embed the metal library - intentional.

Things I have tried:
Running the latest Xcode 16.3
Checked that the iPhoneOS platform is installed under /Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/
Confirmed that the project's build settings look correct for Target:
General - Destination for iPad and Mac(designed for iPad) = iOS
General - Min deployment iOS = 18
Build Settings - Targeted Device Families = iPad
Build Settings - Build Active Architecture Only and Build Active Resources Only = NO

The .sh script builds the iPhoneOS and iPhoneSimulator slices with Device Family set to 2 (iPad). It is modified from the GitHub version of llama.cpp's build-xcframework.sh...

'''
#!/usr/bin/env bash
set -euo pipefail

build-xcframework-metal_dynamic.sh from build-xcframework.sh

Build script for llama.cpp iOS XCFramework (arm64 device + arm64 simulator with Metal)

Run this from the root of the llama.cpp repository.

############### Configuration ################

IOS_DEPLOYMENT_TARGET="16.4" # Minimum iOS version target (adjust as needed)
BUILD_DIR="build" # Output directory for build artifacts

COMMON_C_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
COMMON_CXX_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"

############### Dependency Checks ################

Function to check if a required tool is available

check_tool() {
local tool="$1"
local install_hint="$2"
if ! command -v "$tool" &> /dev/null; then
echo "Error: '$tool' is required but not found. $install_hint"
exit 1
fi
}

Tools to check

check_tool cmake "Please install CMake (e.g., brew install cmake)."
check_tool xcodebuild "Please install Xcode and Xcode Command Line Tools (xcode-select --install)."
check_tool xcrun "Xcode command line tools must be installed for xcrun (metal compiler)."
check_tool libtool "Please install libtool (should be included with Xcode)."
echo "✅ cmake, xcodebuil, xcrun and libtool tools are available."

############### Setup ################

echo "🔧 Cleaning and creating build directories..."
rm -rf "${BUILD_DIR}/ios" "${BUILD_DIR}/llama.xcframework"
mkdir -p "${BUILD_DIR}/ios/device" "${BUILD_DIR}/ios/simulator"
echo "✅ Cleaned and created build directories."

Determine path to the Metal shader source (ggml-metal.metal)

METAL_SRC=""
if [ -f "ggml-metal.metal" ]; then
METAL_SRC="ggml-metal.metal"
elif [ -f "ggml/src/ggml-metal/ggml-metal.metal" ]; then
METAL_SRC="ggml/src/ggml-metal/ggml-metal.metal"
elif [ -f "../ggml/src/ggml-metal/ggml-metal.metal" ]; then
METAL_SRC="../ggml/src/ggml-metal/ggml-metal.metal"]
else
echo "Error: ggml-metal.metal not found in expected location."
exit 1
fi

############### Build for iOS Device (arm64) ################

echo "🔧 Configuring CMake for iOS device (arm64)..."
cmake -S . -B "${BUILD_DIR}/ios/device"
-DCMAKE_SYSTEM_NAME=iOS
-DCMAKE_OSX_DEPLOYMENT_TARGET="${IOS_DEPLOYMENT_TARGET}"
-DCMAKE_OSX_ARCHITECTURES=arm64
-DCMAKE_OSX_SYSROOT=iphoneos
-DCMAKE_BUILD_TYPE=Release
-DCMAKE_C_FLAGS="${COMMON_C_FLAGS}"
-DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}"
-DBUILD_SHARED_LIBS=OFF
-DLLAMA_BUILD_EXAMPLES=OFF
-DLLAMA_BUILD_TESTS=OFF
-DLLAMA_BUILD_SERVER=OFF
-DGGML_METAL=ON
-DGGML_METAL_EMBED_LIBRARY=OFF
-DGGML_METAL_USE_BF16=ON
-DGGML_BLAS_DEFAULT=ON
-DGGML_NATIVE=OFF
-DGGML_ARCH_ARM64=ON
-DDGM_OPENMP=OFF

echo "🔧 Building llama.cpp for iOS device..."
cmake --build "${BUILD_DIR}/ios/device" --config Release || {
echo "❌ Error: Cmake build for device failed."
exit 1
}
echo "✅ Built static libary for iOS device."

############### Build for iOS Simulator (arm64) ################

echo "🔧 Configuring CMake for iOS simulator (arm64)..."
cmake -S . -B "${BUILD_DIR}/ios/simulator"
-DCMAKE_SYSTEM_NAME=iOS
-DCMAKE_OSX_DEPLOYMENT_TARGET="${IOS_DEPLOYMENT_TARGET}"
-DCMAKE_OSX_ARCHITECTURES=arm64
-DCMAKE_OSX_SYSROOT=iphonesimulator
-DCMAKE_BUILD_TYPE=Release
-DCMAKE_C_FLAGS="${COMMON_C_FLAGS}"
-DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}"
-DBUILD_SHARED_LIBS=OFF
-DLLAMA_BUILD_EXAMPLES=OFF
-DLLAMA_BUILD_TESTS=OFF
-DLLAMA_BUILD_SERVER=OFF
-DGGML_METAL=ON
-DGGML_METAL_EMBED_LIBRARY=OFF
-DGGML_METAL_USE_BF16=ON
-DGGML_BLAS_DEFAULT=ON
-DGGML_NATIVE=OFF
-DGGML_ARCH_ARM64=ON
-DDGM_OPENMP=OFF

echo "🔧 Building llama.cpp for iOS simulator..."
cmake --build "${BUILD_DIR}/ios/simulator" --config Release || {
echo "❌ Error: CMake build for simulator failed."
exit 1
}
echo "✅ Built static libary for iOS simulator."

############### Create Universal XCFramework ################

echo "🔧 Combining static libraries from build outputs..."
FRAMEWORK_NAME="llama"
XCFRAMEWORK_OUTPUT_DIR="${BUILD_DIR}/${FRAMEWORK_NAME}.xcframework"
DEVICE_BUILD_DIR="${BUILD_DIR}/ios/device"
SIM_BUILD_DIR="${BUILD_DIR}/ios/simulator"
#DEVICE_LIB="${DEVICE_BUILD_DIR}/src/libllama.a"
#SIM_LIB="${SIM_BUILD_DIR}/src/libllama.a"

#-----------------------------------------------

1. Create Framework structure for device + sim

#-----------------------------------------------
echo "📂 Creating framework folders..."
make_framework() {
local platform=$1 # e.g., ios-arm64 or ios-arm64_x86_64-simulator
local arch=$2 # e.g., arm64
shift 2 # Shift off platform and arch
local lib_paths=("$@") # Remaining arguments are static libraries

local target_dir="${BUILD_DIR}/frameworks/${platform}/${FRAMEWORK_NAME}.framework"

echo "🔧 Creating framework for ${platform}..."
mkdir -p "${target_dir}/Headers"
mkdir -p "${target_dir}/Modules"
echo "✅ Created framework folders"

#---------------------------

Copy all required headers

#---------------------------
echo "📝 Copying header files..."
cp include/llama.h "${target_dir}/Headers"
cp ggml/include/ggml*.h "${target_dir}/Headers"
cp ggml/include/gguf.h "${target_dir}/Headers"
echo "✅ Copied headers."

#-------------------

Create module map

#-------------------
echo "📝 Creating required module map file..."
cat > "${target_dir}/Modules/module.modulemap" <<EOF
framework module ${FRAMEWORK_NAME} {
umbrella header "llama.h"

export *
module * { export * }

link "c++"
link framework "Foundation"
link framework "Accelerate"
link framework "Metal"
}
EOF
echo "✅ Created module map."

#-------------------------------------

Create platform-specific Info.plist

#-------------------------------------
echo "📝 Creating required Info.plist file..."

Determine platform, SDK version, and device family dynamically

local sdk_platform=""
local sdk_version=""
local device_family=""
local cf_platform=""

if [[ "$platform" == simulator ]]; then
sdk_platform="iPhoneSimulator"
echo "🔍 Platform is ${sdk_platform}"
sdk_version=$(xcrun --sdk iphonesimulator --show-sdk-version)
device_family="2" # iPad (iPhone + iPad would be 12)
cf_platform="iPhoneSimulator"
else
sdk_platform="iPhoneOS"
echo "🔍 Platform is ${sdk_platform}"
sdk_version=$(xcrun --sdk iphoneos --show-sdk-version)
device_family="2" # iPad (iPhone + iPad would be 12)
cf_platform="iPhoneOS"
fi

echo "🔍 SDK version is ${sdk_version}"
echo "🔍 Device family is ${device_family} (1 = iPhone, 2 = iPad)"

cat > "${target_dir}/Info.plist" <<EOF

CFBundleDevelopmentRegion en CFBundleExecutable ${FRAMEWORK_NAME} CFBundleIdentifier org.ggml.${FRAMEWORK_NAME} CFBundleInfoDictionaryVersion 6.0 CFBundleName ${FRAMEWORK_NAME} CFBundlePackageType FMWK CFBundleShortVersionString 1.0 CFBundleVersion 1 MinimumOSVersion ${IOS_DEPLOYMENT_TARGET} UIDeviceFamily ${device_family} CFBundleSupportedPlatforms ${cf_platform} DTPlatformName ${cf_platform} DTSDKName ${sdk_platform}${sdk_version} EOF

echo "✅ Created Info.plist."

#--------------------------------------------

2. Create dynamic library for EACH platform

#--------------------------------------------
echo "📦 Creating dynamic library from static libs:"
for lib in "${lib_paths[@]}"; do
echo " 🔗 $lib"
done

Determine correct sysroot and min version for platform

if [[ "$platform" == simulator ]]; then
SDK_PATH=$(xcrun --sdk iphonesimulator --show-sdk-path)
min_version_flag="-mios-simulator-version-min=${IOS_DEPLOYMENT_TARGET}"
else
SDK_PATH=$(xcrun --sdk iphoneos --show-sdk-path)
min_version_flag="-mios-version-min=${IOS_DEPLOYMENT_TARGET}"
fi

echo "🔍 Building for arch: ${arch} on platform: ${platform} for min version ${IOS_DEPLOYMENT_TARGET}"
echo "🔍 Linking with SDK: ${SDK_PATH}"
xcrun clang++ -dynamiclib
-isysroot "$SDK_PATH"
-arch ${arch}
$min_version_flag
"${lib_paths[@]/#/-Wl,-force_load,}"
-framework Foundation -framework Metal -framework Accelerate
-install_name "@rpath/${FRAMEWORK_NAME}.framework/${FRAMEWORK_NAME}"
-o "${target_dir}/${FRAMEWORK_NAME}"
echo "✅ Created dynamic library."

Platform-specific post-processing for device builds - needed to pass App Store Validation of binary

if [[ "$platform" != simulator ]]; then
echo "🔧 Marking binary as a framework binary for iOS..."
if command -v vtool &>/dev/null; then
vtool -set-build-version ios ${IOS_DEPLOYMENT_TARGET} ${IOS_DEPLOYMENT_TARGET} -replace
-output "${target_dir}/${FRAMEWORK_NAME}" "${target_dir}/${FRAMEWORK_NAME}"
echo "✅ Marked bindary with iOS version using vtool."
else
echo "❌ Warning: vtool not found. Binary may not pass App Store validation."
fi
fi
}

#------------------------------------------------------------------------

2. Generate frameworks: make_framework ...

#------------------------------------------------------------------------
make_framework ios-arm64 arm64
"${DEVICE_BUILD_DIR}/src/libllama.a"
"${DEVICE_BUILD_DIR}/ggml/src/libggml.a"
"${DEVICE_BUILD_DIR}/ggml/src/libggml-base.a"
"${DEVICE_BUILD_DIR}/ggml/src/libggml-cpu.a"
"${DEVICE_BUILD_DIR}/ggml/src/ggml-blas/libggml-blas.a"
"${DEVICE_BUILD_DIR}/ggml/src/ggml-metal/libggml-metal.a"
make_framework ios-arm64_x86_64-simulator arm64
"${SIM_BUILD_DIR}/src/libllama.a"
"${SIM_BUILD_DIR}/ggml/src/libggml.a"
"${SIM_BUILD_DIR}/ggml/src/libggml-base.a"
"${SIM_BUILD_DIR}/ggml/src/libggml-cpu.a"
"${SIM_BUILD_DIR}/ggml/src/ggml-blas/libggml-blas.a"
"${SIM_BUILD_DIR}/ggml/src/ggml-metal/libggml-metal.a"

#-----------------------

3. Create XCFramework

#-----------------------
echo "🔧 Creating XCFramework from dynamic frameworks..."
rm -rf "${XCFRAMEWORK_OUTPUT_DIR}"

xcodebuild -create-xcframework
-framework "${BUILD_DIR}/frameworks/ios-arm64/${FRAMEWORK_NAME}.framework"
-framework "${BUILD_DIR}/frameworks/ios-arm64_x86_64-simulator/${FRAMEWORK_NAME}.framework"
-output "${XCFRAMEWORK_OUTPUT_DIR}"

echo "✅ XCFramework created at: ${XCFRAMEWORK_OUTPUT_DIR}"

############### Final Output Information ################

echo ""
echo "🎉 Build complete."
echo "📦 Output XCFramework:"
echo " ${BUILD_DIR}/llama.xcframework"
echo ""
echo "🧠 Output Metal Library:"
echo " ${BUILD_DIR}/ios/device/bin/default.metallib"
echo ""
echo "📱 To integrate into a Swift iOS app:"
echo " 1️⃣ Drag the XCFramework into your Xcode Project Navigator."
echo " 2️⃣ Add 'default.metallib' to your app target's Copy Bundle Resources phase."
echo " 3️⃣ Ensure you link the following system frameworks:"
echo " • Metal"
echo " • Accelerate"
echo " • Foundation"
echo " 4️⃣ ✅ No bridging header needed if you're using 'import llama' — the module map handles that."
echo " 5️⃣ 🔍 At runtime, ensure 'default.metallib' is found in the main bundle (e.g., use Bundle.main.path(forResource: ...))."
echo ""
echo "🚀 Ready to run your model locally using Metal-accelerated inference!"
'''