ggml‑org released llama.cpp version b12362. The update introduces left‑factorization to speed up grammar matching, adds an abstract syntax tree for parsing, and limits alternative rules. Some recent commits were reverted to keep the parser stable. The build now supports a wide range of platforms, from macOS Apple Silicon to Windows CUDA 13.
The release notes list support for Linux, Android, Windows, and various GPU back‑ends such as Vulkan and OpenCL.
Why it matters
Developers can run LLMs with faster, more reliable grammar handling across many devices.
