Aatricks/llmedge-examples

Delivers comprehensive on-device ML inference for Android through the llmedge library, supporting text generation, RAG pipelines, vision models, image/video synthesis, and speech processing (STT/TTS) with GPU acceleration via OpenCL/Vulkan. Uses a modular activity-based architecture demonstrating production patterns for GGUF model loading, ONNX embeddings, streaming inference, and resource management. Integrates Hugging Face Hub for model downloads, Google ML Kit for OCR, and supports popular models like SmolLM, Llava, Stable Diffusion, Wan, Whisper, and Bark.