llama.cpp b9842 fixes duplicate model entries in API

The llama.cpp project, a leading open-source C/C++ inference engine for large language models, shipped version b9842 on June 29. This minor patch addresses a common pain point for developers using its built-in server: duplicate model entries in the /v1/models API endpoint. Previously, presets and cached models could appear multiple times, confusing client applications and tooling. The fix, contributed by Hugging Face engineer Adrien Gallouët, deduplicates both preset and cached entries behind the scenes.

Like all llama.cpp releases, b9842 supports an extensive set of hardware backends — from Apple Silicon (with or without KleidiAI) to Intel, AMD, NVIDIA, and even s390x mainframes. The build matrix includes CPU-only, Vulkan, CUDA 12/13, ROCm, OpenVINO, SYCL, and Adreno for mobile. The release is available as prebuilt binaries for macOS, Linux, Windows, and Android, as well as an iOS XCFramework. No new model features were introduced, making it a straightforward stability update for the server component.