Deploying MNN-LLM to Android devices can be achieved in two ways:
- Pre-compiled APK installation
- Download the Official MNN-LLM Android App from GitHub Releases
- Qwen, Llama, and other models with built-in support are available after installation of APKs
- Source code compilation and deployment
- Clone the project repository: git clone https://github.com/alibaba/MNN.git
- Use Android Studio to open the android/apps/MnnLlmApp directory
- Configure NDK path (r17 and above required)
- Execute Build > Build APK to generate the installation package
Key steps when using:
- Download the target model (e.g. Qwen-7B) within the application, the model file will be stored locally
- Generate responses offline after typing text prompts, support multiple rounds of dialogs
- Multimodal functionality requires image/audio upload via file selector
Note: The current version requires a recompile of the application to load the custom model, and the official promise is that the direct import feature will be opened in the future.
This answer comes from the articleMNN: A Lightweight and Efficient Deep Learning Inference FrameworkThe
May not be reproduced without permission:AI productivity tools " How to deploy MNN-LLM language model on Android devices?
English 
简体中文 
日本語 
Deutsch 
Português do Brasil