The ggml‑org team posted llama.cpp release b11499 on Oct 8 2026. The update changes the CUDA backend to use byte strides for the roll operation, which lets the GPU handle non‑contiguous data correctly. The release also bundles fresh binaries for Apple Silicon, Intel macOS, iOS, and several Ubuntu builds (CPU, Vulkan and CUDA 12). All files are available for download from the GitHub release page.
Why it matters
Developers can run llama.cpp on more GPU setups without errors, expanding local LLM use on desktops and servers.
