The llama.cpp project released version b9915, which introduces support for CPU f16-to-f16 GGML_OP_SET_ROWS operations. This update also includes missing type checks for this operation and merges the f32 and f16 compute forward functions into a single implementation.

  • ggml adds CPU f16->f16 GGML_OP_SET_ROWS support with added type checks
  • ggml_compute_forward_set_rows_f32() and ggml_compute_forward_set_rows_f16() are merged into ggml_compute_forward_set_rows_impl()
  • Assertions are replaced with GGML_ASSERT()
  • macOS Apple Silicon (arm64), macOS Intel (x64), and iOS XCFramework binaries are provided
  • Linux builds include Ubuntu x64, arm64, s390x CPU, Vulkan, ROCm 7.2, OpenVINO, and SYCL FP32/FP16 variants
  • Android arm64 (CPU) binaries are available
  • Windows builds cover x64 and arm64 CPU, OpenCL Adreno, CUDA 12.4 and 13.3, Vulkan, OpenVINO, SYCL, and HIP
  • openEuler supports x86 and aarch64 with ACL Graph for 910b chips

This release provides updated binaries across multiple platforms and hardware accelerators for users requiring the new set rows functionality.