llama.cpp b9976 fixes tensor split warning across platforms
New patch release addresses architecture-specific error message for tensor splitting.
The llama.cpp project, a widely-used C++ implementation for running large language models locally, has released version b9976. This patch addresses a specific issue where a warning message 'LLAMA_SPLIT_MODE_TENSOR not implemented for architecture' was not being displayed correctly under certain conditions. The fix ensures users receive proper feedback when attempting tensor splitting on unsupported architectures, improving debugging and compatibility awareness.
This release also provides comprehensive binary builds across major platforms: macOS (both Apple Silicon and Intel), multiple Linux variants (x64, arm64, s390x with CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Windows (x64, arm64, CUDA 12/13, Vulkan, OpenCL for Adreno, HIP), Android (arm64 CPU), and iOS XCFramework. Notably, the macOS KleidiAI-enabled build and openEuler builds (x86 310p, x86 910b with ACL Graph, aarch64 310p, aarch64 910b with ACL Graph) are disabled in this release. The commit is signed with GitHub's verified signature for security.
- Fixes conditional display of 'LLAMA_SPLIT_MODE_TENSOR not implemented for architecture' warning.
- Provides prebuilt binaries for macOS, Linux, Windows, Android, and iOS across multiple compute backends (CPU, Vulkan, CUDA, ROCm, OpenVINO, SYCL, OpenCL).
- Builds for macOS with KleidiAI and openEuler are disabled in this release.
Why It Matters
Improves error transparency for local LLM users, making tensor splitting behavior clearer across hardware.