onnxruntime-mobile implementation on custom execution provider

[suyash-narain]

to install onnxruntime-mobile, we usualy follow the method here:
https://onnxruntime.ai/docs/install/#install-on-android

and to deploy, we develop android applications as mentioned here:
https://onnxruntime.ai/docs/tutorials/mobile/

In the build.gradle file in android application, we add the implementation as
implementation("com.microsoft.onnxruntime:onnxruntime-android:latest.release")

Now i have a custom execution provider shared lib and libonnxruntime.so which i have built from source. how do i add to my app?

would this be similar to the below implementation?

implementation(files("libs/libonnxruntime.so"))
implementation(files("libs/libexecution_provider.so"))

or we cannot add .so files directly and it needs some other format to add?

[sheetalarkadam]

@suyash-narain, you can place the native files in src/main/native-lib or src/main/jniLibs. The directory can contain subdirectories named after the target architectures (e.g. arm64-v8a,), and you should place the .so files inside the appropriate folder for each architecture.
You don't need to add the .so files explicitly. Gradle will automatically include .so files from the jniLibs folder into your app's APK when it's built.
You could also use .aar files that are generated when you build onnxruntime and use that as the dependency.For reference, you can check the android test in onnxruntime repo implementation(name: "onnxruntime-android", ext: "aar"). Let gradle know local sources by adding it in the repositories