TL;DR: llama.cpp released an update fixing a memory leak in its Metal backend, improving stability and performance for macOS/iOS users.
Summary: ggerganov/llama.cpp, a popular inference engine for large language models, released version b10819. This update specifically addresses and fixes a memory leak issue within its Metal backend, as detailed in pull request #28399.
Why it matters: This fix is crucial for developers deploying LLMs on Apple hardware, ensuring more stable and efficient local inference. Builders should update their llama.cpp installations to leverage these performance improvements on macOS and iOS.
Source: github_releases