The llama.cpp project has released version b9922, which introduces a new `n_keep_tail` parameter in the `split_equal` function specifically for recurrent models.

  • macOS Apple Silicon (arm64) binaries are available.
  • macOS Intel (x64) binaries are provided.
  • iOS XCFramework is included.
  • Ubuntu x64 and arm64 CPU builds are released.
  • Ubuntu Vulkan, ROCm 7.2, OpenVINO, SYCL FP32, and SYCL FP16 variants are available.
  • Android arm64 CPU binaries are provided.
  • Windows x64 and arm64 CPU builds are released.
  • Windows CUDA 12.4, CUDA 13.3, Vulkan, OpenVINO, SYCL, and HIP builds are available.
  • openEuler x86 and aarch64 builds with ACL Graph support are included.
  • A standalone UI package is provided.

This release updates the binary distribution for llama.cpp across multiple platforms and hardware accelerators.