llama.cpp Release b11332 Fixes Recurrent Memory Assert Bug
llama.cpp released patch version b11332, which resolves an invalid assertion bug in recurrent memory (PR #29799). This routine maintenance release prevents unexpected crash assertions during inference on model architectures that utilize recurrent memory state management. The update includes pre-built binaries across macOS, Linux, Windows, and Android, covering hardware backends such as CUDA 12/13, Vulkan, ROCm, OpenVINO, SYCL, and Snapdragon.
## BACKGROUND
llama.cpp is an open-source C/C++ framework for running Large Language Models locally with high performance across varied hardware. Recurrent memory techniques allow language models to process sequences sequentially with linear time complexity and constant memory overhead.