ggerganov/llama.cpp Released Version b11791
Llama.cpp release b11791 includes a bug fix addressing memory leaks and cleanup logic in the Metal GPU capture path. This update improves stability for developers running large language models locally using Apple hardware acceleration. The fix specifically targets issue #30890, refining memory management within the Metal GPU capture implementation.
## BACKGROUND
Llama.cpp is an open-source software library that performs inference on various large language models and serves as a core component for many local AI tools. Metal is Apple's hardware-accelerated graphics and compute API used to leverage Apple Silicon GPUs for efficient model execution.