llama.cpp b12497 adds tiny-voca support
Posted As of
llama.cpp pre-release b12497 adds tiny-voca support to the multimodal path and improves G2P conversion. The release page has no performance figures or description of the target model.
What happened
- Based on GitHub Releases (ggml-org/llama.cpp), this pre-release build was published on 2026-11-18.
- Added tiny-voca support to the mtmd multimodal path (PR #32427).
- Improved G2P (grapheme-to-phoneme) processing.
- Build artifacts include macOS Apple Silicon, Ubuntu, Windows, and Android builds, plus CUDA 13.5 libraries.
- The release page contains no performance figures, benchmarks, or information on which models or users tiny-voca targets.
Our take
The fact that Apple Silicon builds were released alongside this is a useful reference for people running local models on Mac, but the actual effect has not been confirmed. Since this is a pre-release build, its stability also needs to be verified separately.
Sources and related posts
- GitHub Releases (ggml-org/llama.cpp) Primary source