llama.cpp b9621 adds UI fix for original file names and paths

llama.cpp, the ultra-popular open-source C++ inference engine for LLaMA models, has shipped version b9621. The headline change is a UI fix that now keeps original file names and paths when users upload models or files into the interface. Previously, the tool would often rename or flatten paths, causing confusion for developers managing multiple model files. This release also includes a 'fix nocache' tweak that improves memory handling on certain configurations.

On the infrastructure side, the b9621 release offers pre-built binaries for a staggering range of platforms: macOS (Apple Silicon arm64, both standard and KleidiAI-optimized, plus Intel x64 and iOS XCFramework), Linux (x64/arm64/s390x with CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Android (arm64 CPU), and Windows (x64/arm64 CPU, CUDA 12 & 13, Vulkan, SYCL, HIP). This broad compatibility cements llama.cpp as the go-to tool for running LLMs on consumer hardware. With over 116K stars and 19.6K forks on GitHub, the project remains one of the most active in the AI open-source ecosystem, enabling developers to run models like Llama 3 locally without cloud dependencies.