The llama.cpp project has released version b9930, which includes a fix for allowing decreasing positions within a sequence.

  • macOS/iOS: Binaries provided for Apple Silicon (arm64) and Intel (x64), plus an iOS XCFramework. KleidiAI support on Apple Silicon is disabled.
  • Linux: Builds available for Ubuntu x64 and arm64 (CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16).
  • Android: arm64 CPU binary released.
  • Windows: Binaries for x64 and arm64 CPUs, plus OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, and HIP.
  • openEuler: Support disabled for some configurations; builds available for x86 and aarch64 with ACL Graph.
  • A standalone UI binary is also included.