Ggml v1.0.0
|
Ggml v1.0.0 was released on November 13, introducing significant API changes including removal of deprecated functions and legacy operations, plus additions like configurable YaRN truncation and FP8 support. The update brings broad performance improvements across multiple backends including CUDA, Metal, Vulkan, SYCL, OpenCL, WebGPU, and Hexagon through optimizations such as few-row quantized matmuls, MoE fusion, FlashAttention paths, and scheduler improvements that reduce unnecessary data uploads.