b8021

The llama.cpp repository released commit b8021, adding basic OpenCL support for Q4_1 quantization. This enables significantly faster inference of quantized models on AMD GPUs and integrated graphics across Windows, Linux, and macOS. The update includes specific optimizations for matrix-vector and matrix-matrix operations, cleaning up the OpenCL backend. It's part of the ongoing effort to make local LLM inference more accessible and efficient on diverse hardware beyond just NVIDIA CUDA.