b8901

The open-source llama.cpp project, maintained by ggml-org, released version b8901 on April 23, focusing on a critical Metal backend fix. The update addresses issue #22260, which caused event synchronization problems on Apple Silicon (arm64) and iOS devices. This bug could lead to instability during GPU-accelerated LLM inference on macOS and iOS, affecting users running models locally via Metal. The fix ensures proper synchronization of GPU compute events, improving reliability for tasks like text generation and chat with models like Llama 3, Mistral, or Gemma.

The release includes pre-built binaries for multiple platforms: macOS (Apple Silicon and Intel x64), iOS (XCFramework), Linux (x64, arm64, s390x with Vulkan, ROCm 7.2, OpenVINO, SYCL), Windows (x64, arm64 with CUDA 12/13, Vulkan, SYCL, HIP), and Android (arm64). This broad support makes llama.cpp a versatile tool for running LLMs locally across consumer and enterprise hardware. The fix specifically benefits Apple users who rely on Metal for GPU inference, ensuring smoother performance without crashes or hangs. For developers and power users, this update is a minor but crucial improvement for local AI workloads on Apple devices.