Nexa sdk upgrade (#343)

* android sdk upgrade

* update build-aar yml

* update dependencies

* init android audio

* add ttft/tps/decoding speed in android demo app

* update

* update

* updates

* add NexaAudioInference

* upgrade llama.cpp to newest version

* upgrade llama cpp python

* upgrade

* upgrade

* upgrade llama cpp python

* chore: update llama.cpp submodule

* update

* update

* revert some config

* disable bark build

* fix audio issue

* llama cpp to relase branch

* update android cmake omni deps branch

---------

Co-authored-by: Zack Li <[email protected]>
Co-authored-by: JoyboyBrian <[email protected]>
Co-authored-by: Brian <[email protected]>
Co-authored-by: Ubuntu <[email protected]>

.github/workflows/build-aar.yml

@@ -34,15 +34,15 @@ jobs:
     - name: Navigate to android Directory and Build AAR
       run: |
         echo "Navigating to the example directory..."
-        cd android/llama.android
+        cd android
         echo "Starting Gradle build process in $(pwd)..."
         ./gradlew assembleRelease --stacktrace --info
       shell: bash
 
     - name: Rename and upload AAR
       run: |
         echo "Navigating to the android directory to find AAR output..."
-        cd android/llama.android
+        cd android
         mkdir -p ../artifacts
         ls -ld ../artifacts || echo "Artifacts directory does not exist."
         AAR_PATH=$(find ./llama/build/outputs/aar -type f -name "*.aar" | head -n 1)

.gitmodules

@@ -5,7 +5,7 @@
 [submodule "dependency/llama.cpp"]
 	path = dependency/llama.cpp
 	url = https://github.com/NexaAI/llama.cpp.git
-	branch = master-release
+	branch = release
 [submodule "nexa/eval/benchmark_tasks"]
 	path = nexa/eval/benchmark_tasks
 	url = https://github.com/NexaAI/benchmark-tasks.git

android/llama.android/.gitignore → android/.gitignore

@@ -2,6 +2,8 @@
 .gradle/
 build/
 
+.idea
+
 # Local configuration file (sdk path, etc)
 local.properties
 

android/llama.android/README.md → android/README.md

@@ -1,7 +1,7 @@
 # Nexa
 
-**Nexa** offers a convenient Kotlin API for Android developers. It allows seamless integration of llama.cpp models into Android applications.
-**NOTE:** Currently, Nexa supports Large-Language Model (LLM) Vision-Language Model (VLM) inference capabilities.
+**Nexa** is a Kotlin wrapper for the [llama.cpp](https://github.com/ggerganov/llama.cpp.git) library. offering a convenient Kotlin API for Android developers. It allows seamless integration of llama.cpp models into Android applications.
+**NOTE:** Currently, Nexa supports Vision-Language Model (VLM) inference capabilities.
 
 ## Installation
 
@@ -52,3 +52,8 @@ Open the [android test project](./app-java) folder in Android Studio and run the
 ## Download Models
 
 You can download models from the [Nexa AI ModelHub](https://nexa.ai/models).
+
+## How to estimate power usage
+
+- ```adb shell dumpsys batterystats --reset```
+- ```adb shell dumpsys batterystats > batterystats.txt```

android/llama.android/app-java/src/main/AndroidManifest.xml → android/app-java/src/main/AndroidManifest.xml

@@ -1,6 +1,6 @@
 <?xml version="1.0" encoding="utf-8"?>
-<manifest xmlns:android="http://schemas.android.com/apk/res/android">
-
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    xmlns:tools="http://schemas.android.com/tools">
     <uses-permission android:name="android.permission.ACCESS_NETWORK_STATE" />
     <uses-permission android:name="android.permission.INTERNET" />
     <uses-permission android:name="android.permission.RECORD_AUDIO"/>
@@ -12,6 +12,9 @@
     <uses-permission android:name="android.permission.ACCESS_COARSE_LOCATION" />
 
     <application
+        android:largeHeap="true"
+        android:hardwareAccelerated="true"
+        tools:replace="android:largeHeap"
         android:allowBackup="true"
         android:dataExtractionRules="@xml/data_extraction_rules"
         android:fullBackupContent="@xml/backup_rules"