b8679

The open-source project llama.cpp, maintained by ggml-org, has released a significant update with commit b8679. This release introduces two new benchmarking arguments (-fitc and -fitt) to the llama-bench tool, allowing developers to measure inference performance with greater precision across different hardware configurations. The update addresses issue #21304 and includes comprehensive testing across 26 different asset types spanning macOS, Windows, Linux, and openEuler platforms.

The new benchmarking capabilities enable developers to test AI model performance across specialized environments including macOS Apple Silicon (arm64), Windows with CUDA 12.4 and 13.1 DLLs, Linux with Vulkan and ROCm 7.2 support, and openEuler with Huawei Ascend 310p and 910b configurations. This represents one of the most comprehensive cross-platform testing frameworks available for open-source large language models, particularly important for the Llama family of models that llama.cpp supports.

The commit also includes updates to the README.md documentation and the compare-llama-bench.py script, ensuring developers have proper guidance for using the new benchmarking features. This release continues llama.cpp's mission of making large language models accessible and efficient across diverse hardware, from consumer devices to enterprise servers with specialized AI accelerators.