Including multiple kotlin executorch backend libraries is not possible

[vybhav-ibr]

🐛 Describe the bug

I would like to enable three executorch back-ends in my app, I want to use any model compiled for xnnpack and/or vulcan and/or qnn, I am not able to use the libs from maven central,

    implementation("org.pytorch:executorch-android:1.2.0")
    implementation("org.pytorch:executorch-android-qnn:1.2.0")

when both are included , running the app causes the error "Duplicate Class"
I think this is the expected behavior, What is the proper way to do this ?

Versions

.

cc @kirklandsign @cbilgin

[vybhav-ibr]

@cccclai @abhinaykukkadapu
Based on the documentation, to enable multiple back-ends on android, I would need to build the .arr with the appropriate flags.
Would this not cause problems on un-supported hardware, for instance if i build the arr with xnnpack,vulcan, qnn and if the apk is installed on a smartphone without a qualcom chip, would that cause any problems ?

If i only have "org.pytorch:executorch-android-qnn" or "org.pytorch:executorch-android-", would i be able to run a model compiled for xnnpack?, is the xnnpack functionality available in all the back-end libs ?

What is the recommended way to distribute android apps which use executorch, should a different version be distributed for each hardware back-end ?

[kirklandsign]

One workaround is that you can build your own executorch.aar, and use it in your app