Change8

b9802

📦 llama-cppView on GitHub →
18 features🐛 2 fixes

Summary

This release primarily focuses on distributing pre-compiled binaries for a wide array of platforms including macOS, Linux, Android, and Windows, supporting various accelerators like CUDA, ROCm, Vulkan, and OpenVINO.

✨ New Features

  • New binaries provided for macOS Apple Silicon (arm64).
  • New binaries provided for macOS Intel (x64).
  • New iOS XCFramework available.
  • New CPU binaries for Ubuntu (x64, arm64, s390x).
  • New Vulkan binaries for Ubuntu (x64, arm64).
  • New ROCm 7.2 binaries for Ubuntu x64.
  • New OpenVINO binaries for Ubuntu x64 (version 2026.2).
  • New SYCL binaries (FP32 and FP16) for Ubuntu x64.
  • New Android arm64 (CPU) binaries.
  • New Windows CPU binaries (x64 and arm64).
  • New Windows OpenCL Adreno binaries for arm64.
  • New Windows CUDA binaries for versions 12.4 and 13.3 (with separate DLLs provided).
  • New Windows Vulkan binaries (x64).
  • New Windows OpenVINO binaries (version 2026.2, x64).
  • New Windows SYCL binaries (x64).
  • New Windows HIP Radeon binaries (x64).
  • New UI package available.
  • Updates for openEuler builds including 310p and 910b (ACL Graph) support for x86 and aarch64.

🐛 Bug Fixes

  • macOS Apple Silicon (arm64) build with KleidiAI enabled was disabled (referenced PR #23780).
  • openEuler builds were disabled (referenced PR #23705).