llama.cpp Releases Build b11113 with Server Logging Fix
llama.cpp build b11113 has been released, introducing a fix that prevents the server component from passing open log files to child processes. This update addresses file descriptor leaks when spawning child processes within the HTTP server environment. Preventing file descriptors from leaking to child processes improves process isolation, stability, and resource management when hosting local LLM server instances. It ensures background sub-tasks do not accidentally hold open locks or write into the server's primary log file. The fix was implemented via Pull Request #29212 specifically targeting server file handle handling. The release maintains binary builds for macOS, Windows, Linux, and Android across diverse hardware backends including CUDA 12/13, Vulkan, ROCm, OpenVINO, SYCL, and Snapdragon.
## BACKGROUND
llama.cpp is a popular open-source C/C++ LLM inference framework designed to run AI models locally with high efficiency across various CPUs and GPUs. Its built-in HTTP server capability allows users to host local OpenAI-compatible REST APIs, which frequently manage system processes and handles under the hood.