tensorflow
[Android] 16KB Page Sizes support
Android 15 will offer an option for OEMs to use 16KB page sizes [1][2]. According to [1], all shared libraries shipped with Android apps must be recompiled to support 16KB page sizes.
[1] https://developer.android.com/guide/practices/page-sizes [2] https://source.android.com/docs/core/architecture/16kb-page-size/16kb
When will the library get its update?
Same here. Google requires support this for Android 15 otherwise my app will be restricted for updates. Ironic situation when google libraries doesn't met google requirements. For now this issue possibly affects on installations on Android 15 Samsung devices
I've done digging and can confirm the plugin is enforcing -max-page-size=4096
This needs an immediate update, otherwise, a bunch of apps are being forced offline.
@peazz is there any way we can change the page size Enforcing by our own and recompile it for our use case till there is any Official Update ?
Honestly, I am not sure, I done some digging and found this library is the culprit but that is all, I am overwhelmed handling my business alone. right now, I my circle back but this will likely need a update from the repo managers.
This package has not been updated since I started using it and even then, it's not fully functional.
I'm willing to bet that unless this is forked, we won't be getting 16kb page support which would only enable Google to collect on AI costs.
Any update? Because Google play console gives deadline time to 1 November 2025.
Any update or is there any fork we can use, because the deadline is till 1 November 2025
You can make it work if you skip the x86_64 libraries while keeping the ARM versions. Very few phones use them. Add this to build.gradle.kts in the android section:
jniLibs {
excludes.addAll(listOf(
"**/x86_64/**",
"**/x86/**",
"lib/x86_64/**",
"lib/x86/**"
))
// Keep ARM versions of TensorFlow Lite
pickFirsts.addAll(listOf(
"**/arm64-v8a/libtensorflowlite_jni.so",
"**/armeabi-v7a/libtensorflowlite_jni.so",
"**/arm64-v8a/libtensorflowlite_gpu_jni.so",
"**/armeabi-v7a/libtensorflowlite_gpu_jni.so"
))
}
}
and this to the defaultConfig section:
noCompress.add(".tflite")
noCompress.add(".lite")
}
// Only specify ndk filters, no splits
ndk {
abiFilters.clear()
abiFilters.add("armeabi-v7a")
abiFilters.add("arm64-v8a")
// Explicitly exclude x86 architectures
}
This is my buildTypes
release {
signingConfig = signingConfigs.getByName("release")
isMinifyEnabled = true // Re-enable minification
isShrinkResources = true // Re-enable resource shrinking
proguardFiles(
getDefaultProguardFile("proguard-android-optimize.txt"),
"proguard-rules.pro"
)
}
}
where progurard-rules.pro contains
-keep class org.tensorflow.lite.gpu.** { *; }
-dontwarn org.tensorflow.lite.gpu.**
# Additional TensorFlow Lite rules (if needed)
-keep class org.tensorflow.lite.** { *; }
-dontwarn org.tensorflow.lite.**
Now when I inspect the aab in Google Play I see
Native platforms arm64-v8a, armeabi-v7a
API levels 24+
Target SDK 36
OpenGL ES versions 0.0+
OpenGL textures No textures required
Memory page size Supports 16 KB
You can make it work if you skip the x86_64 libraries while keeping the ARM versions. Very few phones use them. Add this to build.gradle.kts in the android section:
jniLibs { excludes.addAll(listOf( "**/x86_64/**", "**/x86/**", "lib/x86_64/**", "lib/x86/**" )) // Keep ARM versions of TensorFlow Lite pickFirsts.addAll(listOf( "**/arm64-v8a/libtensorflowlite_jni.so", "**/armeabi-v7a/libtensorflowlite_jni.so", "**/arm64-v8a/libtensorflowlite_gpu_jni.so", "**/armeabi-v7a/libtensorflowlite_gpu_jni.so" )) } }and this to the defaultConfig section:
noCompress.add(".tflite") noCompress.add(".lite") } // Only specify ndk filters, no splits ndk { abiFilters.clear() abiFilters.add("armeabi-v7a") abiFilters.add("arm64-v8a") // Explicitly exclude x86 architectures }This is my buildTypes
release { signingConfig = signingConfigs.getByName("release") isMinifyEnabled = true // Re-enable minification isShrinkResources = true // Re-enable resource shrinking proguardFiles( getDefaultProguardFile("proguard-android-optimize.txt"), "proguard-rules.pro" ) } }where progurard-rules.pro contains
-keep class org.tensorflow.lite.gpu.** { *; } -dontwarn org.tensorflow.lite.gpu.** # Additional TensorFlow Lite rules (if needed) -keep class org.tensorflow.lite.** { *; } -dontwarn org.tensorflow.lite.**Now when I inspect the aab in Google Play I see
Native platforms arm64-v8a, armeabi-v7a API levels 24+ Target SDK 36 OpenGL ES versions 0.0+ OpenGL textures No textures required Memory page size Supports 16 KB
Hi. It looks like a solution, but arm64-v8a library doesn't support 16KB page sizes either.
The app works on my Samsung Galaxy S24, which uses arm64-v8a. Google says the app supports 16KB, as you can see. So does Profile or Debuge APK in Android Studio. Are you maybe using an old version of the plugin (tensorflow_lite_flutter: ^3.0.0)?
Den fre 26 sep. 2025 kl 08:55 skrev rickgrotavi @.***>:
rickgrotavi left a comment (tensorflow/flutter-tflite#287) https://github.com/tensorflow/flutter-tflite/issues/287#issuecomment-3337037418
You can make it work if you skip the x86_64 libraries while keeping the ARM versions. Very few phones use them. Add this to build.gradle.kts in the android section:
jniLibs { excludes.addAll(listOf( "**/x86_64/**", "**/x86/**", "lib/x86_64/**", "lib/x86/**" )) // Keep ARM versions of TensorFlow Lite pickFirsts.addAll(listOf( "**/arm64-v8a/libtensorflowlite_jni.so", "**/armeabi-v7a/libtensorflowlite_jni.so", "**/arm64-v8a/libtensorflowlite_gpu_jni.so", "**/armeabi-v7a/libtensorflowlite_gpu_jni.so" )) } }and this to the defaultConfig section:
noCompress.add(".tflite") noCompress.add(".lite") } // Only specify ndk filters, no splits ndk { abiFilters.clear() abiFilters.add("armeabi-v7a") abiFilters.add("arm64-v8a") // Explicitly exclude x86 architectures }This is my buildTypes
release { signingConfig = signingConfigs.getByName("release") isMinifyEnabled = true // Re-enable minification isShrinkResources = true // Re-enable resource shrinking proguardFiles( getDefaultProguardFile("proguard-android-optimize.txt"), "proguard-rules.pro" ) } }where progurard-rules.pro contains
-keep class org.tensorflow.lite.gpu.** { ; } -dontwarn org.tensorflow.lite.gpu.*
Additional TensorFlow Lite rules (if needed)
-keep class org.tensorflow.lite.** { ; } -dontwarn org.tensorflow.lite.*
Now when I inspect the aab in Google Play I see
Native platforms arm64-v8a, armeabi-v7a API levels 24+ Target SDK 36 OpenGL ES versions 0.0+ OpenGL textures No textures required Memory page size Supports 16 KB
Hi. It looks like a solution, but arm64-v8a library doesn't support 16KB page sizes either. default.png (view on web) https://github.com/user-attachments/assets/fb7a4b5d-e9f9-44a3-96a6-117400bddb8c
— Reply to this email directly, view it on GitHub https://github.com/tensorflow/flutter-tflite/issues/287#issuecomment-3337037418, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEN5JJIRBIXZYTGP6WKZ6633UTPORAVCNFSM6AAAAACD4A2QCOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGMZXGAZTONBRHA . You are receiving this because you commented.Message ID: @.***>
No, it's not an app issue. The library works fine on every device, but when you upload to the Play Store, they give me a November 1 deadline to fix this. It means the library does not support 16 KB page sizes
The only purpose of the gradle stuff described is to get rid of these Google Play notifications:
To ensure that your app works correctly on the latest versions of Android, Google Play requires all apps targeting Android 15+ to support 16 KB memory page sizes. From 1 Nov 2025, if your app updates do not support 16 KB memory page sizes, you won't be able to release these updates.
For my app, this was caused by the tensorflow_lite_flutter 3.0.0 not supporting 16KB on x86_64. Which should be ok since there are no x86-64 devices with 16kb page size. But Google Play is not intelligent enough to figure out that.
https://github.com/bhanuka96/flutter-tflite/commit/f6e2c80e52f8ccfa3c0ec75cea4d1d8a6af76bba - does this address the issue?
Hi everyone,
I've submitted PR #294 to address 16KB page size support ahead of the Google Play deadline on November 1, 2025.
What it delivers:
- Android Gradle Plugin 8.6.1 (meets Google's requirements)
- SDK compatibility: Android 15+ (API 36)
- All 14 example projects tested and working
- 100% backward compatibility maintained
Testing completed:
- All examples build successfully
- APK generation verified
- Modern Gradle build system implemented
Current status: PR #294 prepares the infrastructure for 16KB support. TensorFlow Lite libraries still need updates from Google's team; this PR provides the foundation to adopt 16KB libraries as soon as they become available.
Notes:
- @BoHellgren's x86 exclusion workaround is useful for immediate needs, but PR #294 provides a comprehensive solution.
- @dewball345: this PR directly addresses the infrastructure requirements for 16KB support.
See PR #294 for technical details.
I’m forking from PR #294 to complete the 16KB page size support. Google has recompiled TensorFlow Lite with 16KB alignment and moved the artifacts from org.tensorflow to com.google.ai.edge.litert.
I’ve updated the dependencies accordingly:
// Old
def tflite_version = "2.12.0"
implementation("org.tensorflow:tensorflow-lite:${tflite_version}")
implementation("org.tensorflow:tensorflow-lite-gpu:${tflite_version}")
// New
def tflite_version = "1.4.0"
implementation("com.google.ai.edge.litert:litert:${tflite_version}")
implementation("com.google.ai.edge.litert:litert-gpu-api:${tflite_version}")
These LiteRT libraries are officially aligned for 16KB page size and fully compatible with existing TensorFlow Lite APIs, so this should resolve the remaining alignment warnings and meet the Google Play requirement for November 2025.
You can test it directly by updating your pubspec.yaml:
tflite_flutter:
git:
url: https://github.com/nguyenvantuyen98/flutter-tflite
ref: main
I have created a PR — please check it @bhanuka96 🙏
@nguyenvantuyen98 your fork compiled fine except X86_64 architecture. For Google Play we can build app without X86_64 temporarily with edited build.gradle config. Add after kotlinOptions
packaging {
jniLibs {
excludes.add("**/x86_64/**")
excludes.add("**/x86/**")
excludes.add("lib/x86_64/**")
excludes.add("lib/x86/**")
}
}
@bhanuka96 Hi, I have updated tflite_flutter dependency to 0.12.0 in my project... I tried uploading an appbundle in Play Store to check if the 16KB page size issue is fixed... But in bundle explorer, i still find this issue...
libraries that do not support 16 kb: base/lib/arm64-v8a/libtensorflowlite_gpu_jni.so base/lib/arm64-v8a/libtensorflowlite_jni.so base/lib/x86_64/libtensorflowlite_c.so base/lib/x86_64/libtensorflowlite_gpu_jni.so base/lib/x86_64/libtensorflowlite_jni.so
Am i missing something? is there anything I need to upgrade or remove in my project??
@NOTtheKRish Oh, thanks for reporting this! Let me check and get back to you.
in a new flutter project i just added this dependency v0.12.0 and ran it on an emulator with system image being pre-release 16kb page size and api 36 and when the project launches its showing this app is not 16kb compatable. @bhanuka96
@NOTtheKRish @VarunMistry88 Please update the dependency to tflite_flutter: 0.12.1 and try again.
@NOTtheKRish @VarunMistry88 Please update the dependency to
tflite_flutter: 0.12.1and try again.
Not working, I have only just added 0.12.1 update to my pubspec, sent it out to Google and it still;
Libraries that do not support 16 KB: base/lib/arm64-v8a/libsqlite3.so base/lib/x86_64/libsqlite3.so
@NOTtheKRish @VarunMistry88 Please update the dependency to
tflite_flutter: 0.12.1and try again.
I've updated to latest version, also removed the jniLabs folder from my project and the appbundle is now supporting 16KB page size (verified with Bundle Explorer in Play Console)
@NOTtheKRish @VarunMistry88 Please update the dependency to
tflite_flutter: 0.12.1and try again.I've updated to latest version, also removed the jniLabs folder from my project and the appbundle is now supporting 16KB page size (verified with Bundle Explorer in Play Console)
I don't have any jniLibs folder and I am on the latest version and bundle explorer is still telling me 16kb is not supported