Dual-backend (Vulkan + OpenCL) prebuilt llama.cpp shared libraries for Android arm64 — a single APK runs vision LLMs on Qualcomm Adreno, ARM Mali, and Samsung Xclipse GPUs with pure-CPU fallback. Built from ggml-org/llama.cpp. Includes mtmd vision + common + llama.cpp headers. No KleidiAI.