Tools
llama.cpp tags its first semver release, v0.5.0, after years of bNNNN build tags
On 23 September ggml-org/llama.cpp published v0.5.0, the only v-prefixed tag among its last 100 releases, which are otherwise bNNNN nightly builds. It ships with structured notes that split out API changes, new models, server and ggml changes. Highlights include CUDA conv2d via implicit GEMM, Metal MoE and SSM_CONV fusion, `input_image` in server function-call outputs, HRM-Text/DFM Mimir 1B support and ggml 0.25.0 with RPC protocol v7. A versioned release gives downstream packagers something stable to pin to instead of chasing daily build numbers.
Source
↳ Follow the thread