llama.cpp b9770 release fixes remote preset handling, adds test

llama.cpp, the lightweight C/C++ implementation for running large language models locally, has released version b9770. This maintenance release focuses on a critical fix to remote preset handling in the server component. Previously, users relying on remote presets (e.g., loading configuration from URLs) could encounter errors that disrupted server operations. The patch resolves this issue and introduces a dedicated test to ensure the fix holds in future iterations. The release is part of the project's continuous improvement cycle, with b9770 compiled and distributed across an extensive list of platforms: macOS (including Apple Silicon with KleidiAI acceleration), Linux (x64, arm64, s390x with Vulkan, ROCm, OpenVINO, SYCL support), Windows (x64, arm64 with CUDA, Vulkan, OpenCL), and Android arm64. The project now boasts over 118,000 stars on GitHub, underscoring its importance in the local AI ecosystem.

For developers and power users running llama.cpp in production or development, this fix ensures more reliable server behavior when using external configuration presets. The addition of tests also improves long-term maintainability. With support for multiple backends (CPU, GPU via CUDA/Vulkan/ROCm, and specialized AI accelerators), llama.cpp remains the go-to tool for deploying LLMs on commodity hardware and edge devices. The b9770 release is available for download from the project's GitHub releases page, with no changes to the core model inference engine—just a targeted bug fix and quality-of-life improvement for server users.