Introducing the new OpenCL™ GPU Backend in llama.cpp for Qualcomm Adreno GPUs

Qualcomm Technologies has announced a new OpenCL-based backend for llama.cpp, optimized for Adreno GPUs in Snapdragon SoCs. This update enhances performance, compatibility, and flexibility, allowing the community to leverage Adreno GPU capabilities for large language models. The backend supports various models including Meta's llama 2 & 3, Gemma, Phi, and Mistral, and has been tested on devices like laptops with Snapdragon X Elite and Android phones with Snapdragon 8 Gen 1, 2, and 3. Detailed steps are provided to build and run llama.cpp on both Android and Windows platforms.