b8732

The llama.cpp project, maintained by the ggml organization, has pushed a critical update with commit b8732. This release specifically addresses a bug in how multimodal padding tokens were handled for Google's recently launched Gemma 3 and Gemma 4 model families. The fix is essential for developers running these models locally, as incorrect token padding can cause inference to fail or produce garbled outputs. The update is now live across all major pre-built binaries.

The patch ensures compatibility across llama.cpp's extensive support matrix, which includes builds for macOS (Apple Silicon and Intel), Linux (CPU, Vulkan, ROCm, OpenVINO), and Windows (CPU, CUDA, Vulkan, SYCL). This wide coverage means the fix applies whether users are running models on a laptop CPU, a desktop GPU with CUDA, or specialized hardware like Intel's OpenVINO or AMD's ROCm platforms. The commit represents a routine but vital maintenance update that keeps the ecosystem stable for cutting-edge model deployment.