llama.cpp Releases Version 0.2.0 with GGML Sync and Kernel Updates

The open-source local LLM inference library llama.cpp has released version 0.2.0. This update synchronizes the project with ggml version 0.21.0 and introduces new kernel support for various hardware backends, including Arm KleidiAI, SYCL, OpenCL, and Vulkan. As a critical library for running large language models locally, official version releases of llama.cpp ensure better stability and performance optimizations across diverse hardware. The integration of optimized kernels like Arm's KleidiAI and Intel's SYCL expands efficient LLM execution on edge devices and heterogeneous systems. Key updates include adding SME2 F32 GEMV kernel support via KleidiAI for Arm architectures, adding Q2_K and Q5_K kernels for SYCL, and implementing workarounds for Adreno GPUs on OpenCL. Additionally, the release introduces CUDA tuning for decoding crossovers and Vulkan optimizations for quantization calculations.