llama.cpp b9753 adds spec model loading progress and stages

llama.cpp version b9753 is now available, marking a targeted improvement to speculative decoding workflows. The update fixes a bug where the server would incorrectly report progress while loading spec (speculative) models, and adds a 'stages' list to clearly indicate each step of the loading process. This is critical for users running speculative decoding—a technique that uses a smaller draft model to speed up inference from a larger target model—as it provides accurate feedback on model preparation.

This release is accompanied by builds for all major platforms, including macOS (Apple Silicon, Intel, iOS), Linux (x64, arm64, s390x with CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL), Windows (CPU, arm64, CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP for AMD), and Android (arm64 CPU). Community contributions include UI assets and nits polishing. For developers and self-hosters running local LLMs with speculative decoding, this fix removes a point of friction and makes monitoring model loading more reliable.