Convert Android LLM extension from Java to Kotlin (#19211)

Summary:
Pull Request resolved: #19211

Convert LlmModule, LlmCallback, LlmGenerationConfig, and LlmModuleConfig
from Java to Kotlin while preserving all public API signatures and JNI
compatibility.

Key changes:
- Use Kotlin idioms: `external fun` for native methods, `companion object`
  for static members, `require()` for validation, `apply {}` in Builders
- Preserve HybridData/fbjni pattern with DoNotStrip and JvmStatic
- Add JvmName annotations for boolean getters (isEcho, isWarming) to
  maintain Java source compatibility
- Keep all constructor overloads as secondary constructors for backward
  compatibility with Java callers (e.g., benchmark LlmModelRunner)
- Update BUCK target language from JAVA to KOTLIN
- Remove package-info.java (no Kotlin equivalent needed)

Differential Revision: D102880053

Author: kirklandsign

extension/android/BUCK

@@ -47,13 +47,13 @@ non_fbcode_target(_kind = fb_android_library,
     name = "executorch_llama",
     warnings_as_errors = False,
     srcs = [
-        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java",
-        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmGenerationConfig.java",
-        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.java",
-        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModuleConfig.java",
+        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.kt",
+        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmGenerationConfig.kt",
+        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModule.kt",
+        "executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmModuleConfig.kt",
     ],
     autoglob = False,
-    language = "JAVA",
+    language = "KOTLIN",
     deps = [
         ":executorch",
         "//fbandroid/java/com/facebook/jni:jni",

…xecutorch/extension/llm/LlmCallback.java …/executorch/extension/llm/LlmCallback.ktextension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java renamed to extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.kt extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.java renamed to extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmCallback.kt

@@ -6,45 +6,42 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-package org.pytorch.executorch.extension.llm;
+package org.pytorch.executorch.extension.llm
 
-import com.facebook.jni.annotations.DoNotStrip;
-import org.pytorch.executorch.annotations.Experimental;
+import com.facebook.jni.annotations.DoNotStrip
+import org.pytorch.executorch.annotations.Experimental
 
 /**
- * Callback interface for Llama model. Users can implement this interface to receive the generated
+ * Callback interface for Llm model. Users can implement this interface to receive the generated
  * tokens and statistics.
  *
- * <p>Warning: These APIs are experimental and subject to change without notice
+ * Warning: These APIs are experimental and subject to change without notice
  */
 @Experimental
-public interface LlmCallback {
+interface LlmCallback {
   /**
    * Called when a new result is available from JNI. Users will keep getting onResult() invocations
    * until generate() finishes.
    *
    * @param result Last generated token
    */
-  @DoNotStrip
-  public void onResult(String result);
+  @DoNotStrip fun onResult(result: String)
 
   /**
    * Called when the statistics for the generate() is available.
    *
-   * <p>The result will be a JSON string. See extension/llm/stats.h for the field definitions.
+   * The result will be a JSON string. See extension/llm/stats.h for the field definitions.
    *
    * @param stats JSON string containing the statistics for the generate()
    */
-  @DoNotStrip
-  default void onStats(String stats) {}
+  @DoNotStrip fun onStats(stats: String) {}
 
   /**
    * Called when an error occurs during generate().
    *
-   * @param errorCode Error code from the ExecuTorch runtime (see {@link
-   *     org.pytorch.executorch.ExecutorchRuntimeException})
+   * @param errorCode Error code from the ExecuTorch runtime (see
+   *   [org.pytorch.executorch.ExecutorchRuntimeException])
    * @param message Human-readable error description
    */
-  @DoNotStrip
-  default void onError(int errorCode, String message) {}
+  @DoNotStrip fun onError(errorCode: Int, message: String) {}
 }

extension/android/executorch_android/src/main/java/org/pytorch/executorch/extension/llm/LlmGenerationConfig.java

@@ -0,0 +1,78 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+package org.pytorch.executorch.extension.llm
+
+/**
+ * Configuration class for controlling text generation parameters in LLM operations.
+ *
+ * This class provides settings for text generation behavior including output formatting, generation
+ * limits, and sampling parameters. Instances should be created using the [create] method and the
+ * fluent builder pattern.
+ */
+class LlmGenerationConfig
+private constructor(
+    @get:JvmName("isEcho") val echo: Boolean,
+    val maxNewTokens: Int,
+    @get:JvmName("isWarming") val warming: Boolean,
+    val seqLen: Int,
+    val temperature: Float,
+    val numBos: Int,
+    val numEos: Int,
+) {
+
+  companion object {
+    /**
+     * Creates a new Builder instance for constructing generation configurations.
+     *
+     * @return a new Builder with default configuration values
+     */
+    @JvmStatic fun create(): Builder = Builder()
+  }
+
+  /**
+   * Builder class for constructing LlmGenerationConfig instances.
+   *
+   * Provides a fluent interface for configuring generation parameters with sensible defaults. All
+   * methods return the builder instance to enable method chaining.
+   */
+  class Builder internal constructor() {
+    private var echo: Boolean = true
+    private var maxNewTokens: Int = -1
+    private var warming: Boolean = false
+    private var seqLen: Int = -1
+    private var temperature: Float = 0.8f
+    private var numBos: Int = 0
+    private var numEos: Int = 0
+
+    /** Sets whether to include the input prompt in the generated output. */
+    fun echo(echo: Boolean): Builder = apply { this.echo = echo }
+
+    /** Sets the maximum number of new tokens to generate. */
+    fun maxNewTokens(maxNewTokens: Int): Builder = apply { this.maxNewTokens = maxNewTokens }
+
+    /** Enables or disables model warming. */
+    fun warming(warming: Boolean): Builder = apply { this.warming = warming }
+
+    /** Sets the maximum sequence length for generation. */
+    fun seqLen(seqLen: Int): Builder = apply { this.seqLen = seqLen }
+
+    /** Sets the temperature for random sampling. */
+    fun temperature(temperature: Float): Builder = apply { this.temperature = temperature }
+
+    /** Sets the number of BOS tokens to prepend. */
+    fun numBos(numBos: Int): Builder = apply { this.numBos = numBos }
+
+    /** Sets the number of EOS tokens to append. */
+    fun numEos(numEos: Int): Builder = apply { this.numEos = numEos }
+
+    /** Constructs the LlmGenerationConfig instance with the configured parameters. */
+    fun build(): LlmGenerationConfig =
+        LlmGenerationConfig(echo, maxNewTokens, warming, seqLen, temperature, numBos, numEos)
+  }
+}