Installation

Previous Next

This guide covers installing and building llmedge for local development and integrating the library into an Android project.

Prerequisites

  • JDK 17+ (for Gradle Kotlin DSL)
  • Android SDK & NDK (if building Android native parts)
  • CMake and a C++ toolchain supporting your target (clang/gcc)
  • Python 3.8+ for some auxiliary scripts (optional)
  • (Optional) Conda/venv for any Python tooling

Project layout

  • llmedge/ — main Android library module with JNI/C++ sources
  • llmedge-examples/app/ — Android sample app showing use-cases
  • llama.cpp/ — vendored/third-party inference code used by native layer

For simple usage, you can download prebuilt AARs from the Releases page and include them in your Android project.

For development and building from source, follow these steps: Clone

git clone https://github.com/Aatricks/llmedge.git
cd llmedge
git submodule update --init --recursive

Build (Android Studio)

  1. Open the llmedge root in Android Studio (or the llmedge-examples project).
  2. Let Gradle sync. Android Studio will download required SDK/NDK and Gradle toolchains per local.properties and gradle.properties.
  3. Build the :llmedge library and example app (Build -> Make Project).

Command-line build

  • Build AAR of llmedge (from project root):
./gradlew :llmedge:assembleDebug
  • Build the example app:
./gradlew :llmedge-examples:app:assembleDebug

Native build notes

  • The native code resides in llmedge/src/main/cpp and uses JNI bindings and llama.cpp-style readers.
  • If you run into issues with the C++ toolchain, verify ANDROID_NDK_HOME and ANDROID_SDK_ROOT in local.properties or environment variables.

GGUF/Model files

  • GGUFReader supports loading GGUF model files. Place model files on device storage or in the app's files directory and point APIs at those paths.

Hugging Face integration

  • The io.aatricks.llmedge.huggingface package can download models from the Hugging Face Hub.
  • Use SmolLM.loadFromHuggingFace() for automatic download and loading.
  • Optionally provide an HF token for private repositories via the token parameter.

Troubleshooting

  • Gradle sync failed: delete .gradle and try again
  • Native build fails: ensure correct NDK version and API level
  • Missing model: confirm the model path and file permissions on device

If you prefer a containerized CI build, refer to llama.cpp/ci for example scripts that set up required toolchains.

Home Usage