llama.cpp b9832 adds Jinja template debugging with --dump-prog

llama.cpp, the widely-used open-source C++ library for running LLaMA and other large language models locally, has released version b9832. The key addition is a new command-line flag `--dump-prog` for Jinja template debugging. Jinja templates are commonly used in LLM inference to format chat prompts and system messages. This flag allows developers to inspect the compiled Jinja program, making it easier to debug custom prompt templates and ensure correct formatting before inference. The feature is implemented in the `common/jinja/runtime.cpp` file and was contributed with a verified GPG signature.

The release also showcases llama.cpp's extensive cross-platform support. Builds are provided for macOS (Apple Silicon arm64, Intel x64, and iOS XCFramework), Linux (x64 and arm64 with CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL), Windows (x64 and arm64 with CPU, CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP), and Android arm64. Notably, some builds are disabled on openEuler. This breadth ensures that developers can run local LLM inference on nearly any modern hardware, from consumer laptops to high-end GPU servers. Version b9832 continues llama.cpp's trajectory as the go-to tool for private, efficient AI inference.