Phi-3-mini is a new series of models from Microsoft that enables deployment of Large Language Models (LLMs) on edge devices and IoT devices. Phi-3-mini is available for iOS, Android, and Edge Device deployments, allowing generative AI to be deployed in BYOD environments. The following example demonstrates how to deploy Phi-3-mini on iOS.

• a. macOS 14+
• b. Xcode 15+
• c. iOS SDK 17.x (iPhone 14 A16 or greater)
• d. Install Python 3.10+ (Conda is recommended)
• e. Install the Python library: python-flatbuffers
• f. Install CMake

Semantic Kernel is an application framework that allows you to create applications compatible with Azure OpenAI Service, OpenAI models, and even local models. Accessing local services through Semantic Kernel enables easy integration with your self-hosted Phi-3-mini model server.

Many users prefer using quantized models to run models locally. Ollama and LlamaEdge allow users to call different quantized models:

You can run ollama run phi3 directly or configure it offline. Create a Modelfile with the path to your gguf file. Sample code for running the Phi-3-mini quantized model:

FROM {Add your gguf file path}
TEMPLATE \"\"\"<|user|> .Prompt<|end|> <|assistant|>\"\"\"
PARAMETER stop <|end|>
PARAMETER num_ctx 4096

If you want to use gguf in both cloud and edge devices simultaneously, LlamaEdge is a great option.

git clone https://github.com/microsoft/onnxruntime.git

cd onnxruntime

./build.sh --build_shared_lib --ios --skip_tests --parallel --build_dir ./build_ios --ios --apple_sysroot iphoneos --osx_arch arm64 --apple_deploy_target 17.5 --cmake_generator Xcode --config Release

cd ../

• a. Before compiling, ensure that Xcode is properly configured and set it as the active developer directory in the terminal:

  sudo xcode-select -switch /Applications/Xcode.app/Contents/Developer

• b. ONNX Runtime needs to be compiled for different platforms. For iOS, you can compile for arm64 or x86_64.

• c. It is recommended to use the latest iOS SDK for compilation. However, you can also use an older version if you need compatibility with previous SDKs.

Note: Because Generative AI with ONNX Runtime is in preview, please be aware of potential changes.

git clone https://github.com/microsoft/onnxruntime-genai
 
cd onnxruntime-genai
 
mkdir ort
 
cd ort
 
mkdir include
 
mkdir lib
 
cd ../
 
cp ../onnxruntime/include/onnxruntime/core/session/onnxruntime_c_api.h ort/include
 
cp ../onnxruntime/build_ios/Release/Release-iphoneos/libonnxruntime*.dylib* ort/lib
 
export OPENCV_SKIP_XCODEBUILD_FORCE_TRYCOMPILE_DEBUG=1
 
python3 build.py --parallel --build_dir ./build_ios --ios --ios_sysroot iphoneos --ios_arch arm64 --ios_deployment_target 17.5 --cmake_generator Xcode --cmake_extra_defines CMAKE_XCODE_ATTRIBUTE_CODE_SIGNING_ALLOWED=NO

I chose Objective-C as the App development method , because using Generative AI with ONNX Runtime C++ API, Objective-C is better compatible. Of course, you can also complete related calls through Swift bridging.

We need to import the INT4 quantization model in ONNX format, which needs to be downloaded first

After downloading, you need to add it to the Resources directory of the project in Xcode.

Notice:

• a. Add the corresponding C++ header files to the project.

  

• b. Include the onnxruntime-genai dynamic library in Xcode.

  

• c. Use the C Samples code for testing. You can also add additional features like ChatUI for more functionality.

• d. Since you need to use C++ in your project, rename ViewController.m to ViewController.mm to enable Objective-C++ support.

    NSString *llmPath = [[NSBundle mainBundle] resourcePath];
    char const *modelPath = llmPath.cString;

    auto model =  OgaModel::Create(modelPath);

    auto tokenizer = OgaTokenizer::Create(*model);

    const char* prompt = "<|system|>You are a helpful AI assistant.<|end|><|user|>Can you introduce yourself?<|end|><|assistant|>";

    auto sequences = OgaSequences::Create();
    tokenizer->Encode(prompt, *sequences);

    auto params = OgaGeneratorParams::Create(*model);
    params->SetSearchOption("max_length", 100);
    params->SetInputSequences(*sequences);

    auto output_sequences = model->Generate(*params);
    const auto output_sequence_length = output_sequences->SequenceCount(0);
    const auto* output_sequence_data = output_sequences->SequenceData(0);
    auto out_string = tokenizer->Decode(output_sequence_data, output_sequence_length);
    
    auto tmp = out_string;

Once the setup is complete, you can run the application to see the results of the Phi-3-mini model inference.

For more sample code and detailed instructions, visit the Phi-3 Mini Samples repository.