core: swap out hardcoded LlamaAndroid library loading

examples/llama.android/app/src/main/java/com/example/llama/legacy/LegacyViewModel.kt

@@ -1,6 +1,8 @@
 package com.example.llama.legacy
 
+import android.content.Context
 import android.llama.cpp.LLamaAndroid
+import android.llama.cpp.LLamaLibraryLoader
 import android.util.Log
 import androidx.compose.runtime.getValue
 import androidx.compose.runtime.mutableStateOf
@@ -10,15 +12,16 @@ import androidx.lifecycle.viewModelScope
 import kotlinx.coroutines.flow.catch
 import kotlinx.coroutines.launch
 
-class LegacyViewModel(
-    private val llamaAndroid: LLamaAndroid = LLamaAndroid.instance()
-): ViewModel() {
+class LegacyViewModel(context: Context): ViewModel() {
     companion object {
+        private val tag = LegacyViewModel::class.java.simpleName
+
         @JvmStatic
         private val NanosPerSecond = 1_000_000_000.0
     }
 
-    private val tag: String? = this::class.simpleName
+    val llamaAndroid: LLamaAndroid = LLamaLibraryLoader.createInstance(context)
+        ?: throw InstantiationException("Cannot instantiate LlamaAndroid!")
 
     var messages by mutableStateOf(listOf("Initializing..."))
         private set

examples/llama.android/app/src/main/java/com/example/llama/revamp/di/AppModule.kt

@@ -2,7 +2,7 @@ package com.example.llama.revamp.di
 
 import android.content.Context
 import android.llama.cpp.InferenceEngine
-import android.llama.cpp.LLamaAndroid
+import android.llama.cpp.LLamaLibraryLoader
 import com.example.llama.revamp.data.local.AppDatabase
 import com.example.llama.revamp.data.remote.HuggingFaceApiService
 import com.example.llama.revamp.data.remote.HuggingFaceRemoteDataSource
@@ -58,10 +58,15 @@ internal abstract class AppModule {
     ): HuggingFaceRemoteDataSource
 
     companion object {
+        private const val USE_REAL_ENGINE = true
+
         @Provides
-        fun provideInferenceEngine(): InferenceEngine {
-            val useRealEngine = true
-            return if (useRealEngine) LLamaAndroid.instance() else StubInferenceEngine()
+        fun provideInferenceEngine(@ApplicationContext context: Context): InferenceEngine {
+            return if (USE_REAL_ENGINE) {
+                LLamaLibraryLoader.createInstance(context) ?: throw InstantiationException("Cannot instantiate LlamaAndroid!")
+            } else {
+                StubInferenceEngine()
+            }
         }
 
         @Provides

examples/llama.android/app/src/main/java/com/example/llama/revamp/engine/StubInferenceEngine.kt

@@ -16,15 +16,11 @@ import kotlinx.coroutines.flow.catch
 import kotlinx.coroutines.flow.flow
 import kotlinx.coroutines.launch
 import kotlinx.coroutines.withContext
-import org.jetbrains.annotations.TestOnly
-import org.jetbrains.annotations.VisibleForTesting
 import javax.inject.Singleton
 
 /**
  * A stub [InferenceEngine] for agile development & testing
  */
-@VisibleForTesting
-@TestOnly
 @Singleton
 class StubInferenceEngine : InferenceEngine {
     companion object {

examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt

@@ -1,7 +1,6 @@
 package android.llama.cpp
 
 import android.llama.cpp.InferenceEngine.State
-import android.llama.cpp.LLamaAndroid.Companion.instance
 import android.util.Log
 import kotlinx.coroutines.CancellationException
 import kotlinx.coroutines.CoroutineScope
@@ -37,7 +36,35 @@ import java.io.File
  *
  * @see llama-android.cpp for the native implementation details
  */
-class LLamaAndroid private constructor() : InferenceEngine {
+class LLamaAndroid private constructor(private val tier: LLamaTier) : InferenceEngine {
+
+    companion object {
+        private val TAG = LLamaAndroid::class.java.simpleName
+
+        private var initialized = false
+
+        /**
+         * Create LLamaAndroid instance with specific tier
+         */
+        internal fun createWithTier(tier: LLamaTier): LLamaAndroid? {
+            if (initialized) {
+                Log.w(TAG, "LLamaAndroid already initialized")
+                return null
+            }
+
+            try {
+                Log.i(TAG, "Instantiating LLamaAndroid w/ ${tier.libraryName}")
+                val instance = LLamaAndroid(tier)
+                initialized = true
+                return instance
+
+            } catch (e: UnsatisfiedLinkError) {
+                Log.e(TAG, "Failed to load ${tier.libraryName}", e)
+                return null
+            }
+        }
+    }
+
     /**
      * JNI methods
      * @see llama-android.cpp
@@ -74,13 +101,14 @@ class LLamaAndroid private constructor() : InferenceEngine {
                 check(_state.value is State.Uninitialized) {
                     "Cannot load native library in ${_state.value.javaClass.simpleName}!"
                 }
-
                 _state.value = State.Initializing
-                Log.i(TAG, "Loading native library $LIB_LLAMA_ANDROID")
-                System.loadLibrary(LIB_LLAMA_ANDROID)
+                Log.i(TAG, "Loading native library for $tier")
+
+                System.loadLibrary(tier.libraryName)
                 init()
                 _state.value = State.Initialized
                 Log.i(TAG, "Native library loaded! System info: \n${systemInfo()}")
+
             } catch (e: Exception) {
                 Log.e(TAG, "Failed to load native library", e)
                 throw e
@@ -235,15 +263,4 @@ class LLamaAndroid private constructor() : InferenceEngine {
             else -> { unload(); shutdown() }
         }
     }
-
-    companion object {
-        private val TAG = LLamaAndroid::class.simpleName
-
-        // TODO-han.yin: replace with dynamic loader
-        private const val LIB_LLAMA_ANDROID = "llama_android_t3"
-
-        // Enforce only one instance of Llm.
-        private val _instance: LLamaAndroid = LLamaAndroid()
-        fun instance(): LLamaAndroid = _instance
-    }
 }