On-Device Deployment

The following describes how to deploy the Federated-Client in the Android aarch and Linux x86_64 environments:

Android aarch

Building a Package

Configure the build environment.

Currently, only the Linux build environment is supported. For details about how to configure the Linux build environment, click here.
Build the x86-related architecture package in the mindspore home directory.
```
bash build.sh -I x86_64 -j32
```
And the x86 architecture package will be generated in the path mindspore/output/after compiling ( please backup it to avoid auto-deletion while next compile):
```
mindspore-lite-{version}-linux-x64.tar.gz
```
Turn on Federated-Client compile option and build the AAR package that contains aarch64 and aarch32 in the mindspore home directory.
```
export MSLITE_ENABLE_FL=on
bash build.sh -A on -j32
```
The Android AAR package will be generated in the path mindspore/output/ after compiling ( please backup it to avoid auto-deletion while next compile):
```
mindspore-lite-full-{version}.aar
```
Since the device-side framework and the model are decoupled, we provide Android AAR package mindspore-lite-full-{version}.aar that does not contain model-related scripts, so users need to generate the model script corresponding to the jar package. We provide two types of model scripts for your reference (Supervised sentiment Classification Task, LeNet image classification task). Users can refer to these two types of model scripts, and generate the corresponding jar package (assuming the name is quick_start_flclient.jar) after customizing the model script. The jar packages corresponding to the model scripts we provide can be obtained in the following ways:

After downloading the latest code on MindSpore Open Source Warehouse, perform the following operations:
```
cd mindspore/mindspore/lite/examples/quick_start_flclient
sh build.sh -r "mindspore-lite-{version}-linux-x64.tar.gz"   # After -r, give the absolute path of the latest x86 architecture package that generate at step 2
```
After running the above command, the path of the jar package generated is: mindspore/mindspore/lite/examples/quick_start_flclient/target/quick_start_flclient.jar.

Running Dependencies

Android Studio >= 4.0
Android SDK >= 29

Building a Dependency Environment

Renaming mindspore-lite-full-{version}.aar to mindspore-lite-full-{version}.zip. After the mindspore-lite-full-{version}.zip file is decompressed, the following directory structure is obtained:

mindspore-lite-full-{version}
├── jni
│   ├── arm64-v8a
│   │   ├── libjpeg.so   # Dynamic library file for image processing
│   │   ├── libminddata-lite.so  # Dynamic library file for image processing
│   │   ├── libmindspore-lite.so  # Dynamic library on which the MindSpore Lite inference framework depends
│   │   ├── libmindspore-lite-jni.so  # JNI dynamic library on which the MindSpore Lite inference framework depends
│   │   ├── libmindspore-lite-train.so  # Dynamic library on which the MindSpore Lite training framework depends
│   │   ├── libmindspore-lite-train-jni.so  # JNI dynamic library on which the MindSpore Lite training framework depends
│   │   └── libturbojpeg.so  # Dynamic library file for image processing
│   └── armeabi-v7a
 │       ├── libjpeg.so   # Dynamic library file for image processing
│       ├── libminddata-lite.so  # Dynamic library file for image processing
│       ├── libmindspore-lite.so  # Dynamic library on which the MindSpore Lite inference framework depends
│       ├── libmindspore-lite-jni.so  # JNI dynamic library on which the MindSpore Lite inference framework depends
│       ├── libmindspore-lite-train.so  # Dynamic library on which the MindSpore Lite training framework depends
│       ├── libmindspore-lite-train-jni.so  # JNI dynamic library on which the MindSpore Lite training framework depends
│       └── libturbojpeg.so  # Dynamic library file for image processing
├── libs
│   ├── mindspore-lite-java-common.jar  # MindSpore Lite training framework JAR package
│   └── mindspore-lite-java-flclient.jar  # Federated learning framework JAR package
└── classes.jar  # MindSpore Lite training framework JAR package

Note 1: since the federated learning jar package in the Android environment does not contain the dependent third-party open source software packages, in the Android environment, before using the AAR package, the user needs to add related dependency statements in the dependencies{} field to load the three open source software that Federated Learning depends on, and the dependencies{} field is in the app/build.gradle file under the Android project, as shown below:

dependencies {

// Add third-party open source software that federated learning relies on
implementation group: 'com.squareup.okhttp3', name: 'okhttp', version: '3.14.9'
implementation group: 'com.google.flatbuffers', name: 'flatbuffers-java', version: '2.0.0'
implementation(group: 'org.bouncycastle',name: 'bcprov-jdk15on', version: '1.68')
}

For specific implementation, please refer to the example of app/build.gradle provided in the Android project configuration dependencies section in the document sentiment classification application.

Note 2: since the third-party open source software bcprov-jdk15on that Federated Learning relies on contains multi-version class files, in order to prevent errors in compiling high-version class files with lower version jdk, the following setting statement can be added to the gradle.properties file of the Android project:

android.jetifier.blacklist=bcprov

After setting up the dependencies shown above in the Android project, you only need to rely on the AAR package and the jar package corresponding to the model script quick_start_flclient.jar to call APIs provided by federated learning. For details about how to call and run the APIs, see the API description of federated learning.

Linux x86_64

Building a Package

Configure the build environment.

Currently, only the Linux build environment is supported. For details about how to configure the Linux build environment, click here.
Build the x86-related architecture package in the mindspore home directory
```
bash build.sh -I x86_64 -j32
```
And the x86 architecture package will be generated in the path mindspore/output/ after compiling ( please backup it to avoid auto-deletion while next compile):
```
mindspore-lite-{version}-linux-x64.tar.gz
```
Since the device-side framework and the model are decoupled, we provide x86 architecture package mindspore-lite-{version}-linux-x64.tar.gz that does not contain model-related scripts, so users need to generate the model script corresponding to the jar package. We provide two types of model scripts for your reference (Supervised sentiment Classification Task, LeNet image classification task). Users can refer to these two types of model scripts, and generate the corresponding jar package (assuming the name is quick_start_flclient.jar) after customizing the model script. The jar packages corresponding to the model scripts we provide can be obtained in the following ways:

After downloading the latest code on MindSpore Open Source Warehouse, perform the following operations:
```
cd mindspore/mindspore/lite/examples/quick_start_flclient
sh build.sh -r "mindspore-lite-{version}-linux-x64.tar.gz" # After -r, give the absolute path of the latest x86 architecture package that generate at step 2
```
After running the above command, the path of the jar package generated is: mindspore/mindspore/lite/examples/quick_start_flclient/target/quick_start_flclient.jar.

Running Dependencies

Python >= 3.7.0
OpenJDK 1.8 to 1.15

Building a Dependency Environment

After the mindspore/output/mindspore-lite-{version}-linux-x64.tar.gz file is decompressed, the following directory structure is obtained:

mindspore-lite-{version}-linux-x64
├── tools
│   ├── benchmark_train # Tool for commissioning the performance and accuracy of the training model
│   ├── converter       # Model conversion tool
│   └── cropper         # Library cropping tool
│       ├── cropper                 # Executable file of the library cropping tool
│       └── cropper_mapping_cpu.cfg # Configuration file required for cropping the CPU library
└── runtime
    ├── include  # Header file of the training framework
    │   └── registry # Header file for custom operator registration
    ├── lib      # Training framework library
    │   ├── libminddata-lite.a          # Static library file for image processing
    │   ├── libminddata-lite.so        # Dynamic library file for image processing
    │   ├── libmindspore-lite-jni.so   # JNI dynamic library on which the MindSpore Lite inference framework depends
    │   ├── libmindspore-lite-train.a  # Static library on which the MindSpore Lite training framework depends
    │   ├── libmindspore-lite-train.so # Dynamic library on which the MindSpore Lite training framework depends
    │   ├── libmindspore-lite-train-jni.so # JNI dynamic library on which the MindSpore Lite training framework depends
    │   ├── libmindspore-lite.a  # Static library on which the MindSpore Lite inference framework depends
    │   ├── libmindspore-lite.so  # Dynamic library on which the MindSpore Lite inference framework depends
    │   ├── mindspore-lite-java.jar    # MindSpore Lite training framework JAR package
    │   └── mindspore-lite-java-flclient.jar  # Federated learning framework JAR package
    └── third_party
        └── libjpeg-turbo
            └── lib
                ├── libjpeg.so.62   # Dynamic library file for image processing
                └── libturbojpeg.so.0  # Dynamic library file for image processing

The x86 packages required for federated learning are as follows:

libjpeg.so.62   # Dynamic library file for image processing
libminddata-lite.so  # Dynamic library file for image processing
libmindspore-lite.so  # Dynamic library on which the MindSpore Lite inference framework depends
libmindspore-lite-jni.so  # JNI dynamic library on which the MindSpore Lite inference framework depends
libmindspore-lite-train.so  # Dynamic library on which the MindSpore Lite training framework depends
libmindspore-lite-train-jni.so # JNI dynamic library on which the MindSpore Lite training framework depends
libturbojpeg.so.0  # Dynamic library file for image processing
mindspore-lite-java-flclient.jar  # Federated learning framework JAR package
quick_start_flclient.jar  # The jar package corresponding to the model script

Find the seven .so files on which federated learning depends in the directories mindspore/output/mindspore-lite-{version}-linux-x64/runtime/lib/ and mindspore/output/mindspore-lite-{version}-linux-x64/runtime/third_party/libjpeg-turbo/lib. Then, place these .so files in a folder, for example, /resource/x86libs/.

Set environment variables in the x86 system (an absolute path must be provided):

export LD_LIBRARY_PATH=/resource/x86libs/:$LD_LIBRARY_PATH

After the dependency environment is set, you can simulate the startup of multiple clients in the x86 environment for federated learning. For details, click here.