llama.cpp v0.5.0 released — model, backend and server updates
Release v0.5.0 updates backends and model coverage: HRM-Text (DFM Mimir 1B), MiMo-V2.6, HunyuanOCR support, ggml v0.25.0 backend changes, multi-address HTTP binding, image outputs from function calls, and UI/server fixes.
What changed
Added HRM-Text (DFM Mimir 1B), MiMo-V2.6 and HunyuanOCR conversions; updated ggml to v0.25.0; allowed server binding to multiple addresses; added image outputs from function calls; CUDA/Metal backend optimizations.
Why it matters
Developers and operators can support more model formats and benefit from backend correctness/optimizations; operators get more flexible network binding; creators can receive image outputs from function calls.
Who should care
DevelopersCreatorsEnterprises
Worth trying
Yes — Use for new model support and ggml v0.25.0 backend improvements.