ggerganov/llama.cpp Released b12116 with Drama-3 TTS Support
Release b12116 of llama.cpp introduces Drama-3 text-to-speech support via a new multi-modal model feature. It also provides standard pre-built multi-platform binary packages across macOS, Linux, Windows, and Android. This release expands the library's capabilities beyond large language models into audio synthesis, enhancing multi-modal deployment options for local AI developers. It reinforces llama.cpp's role as a comprehensive infrastructure backbone for running various neural network models efficiently on edge devices. The update includes pull request #31613, which implements the Drama-3 text-to-speech helper logic with a safeguard that stops processing at the codebook tower limit. Additionally, the release bundles multiple specialized hardware builds such as CUDA, Vulkan, OpenVINO, and Snapdragon-optimized variants.
## BACKGROUND
llama.cpp is an open-source software library written in C/C++ that enables efficient local inference for large language models and other machine learning architectures. Co-developed alongside the GGML tensor library, it has become a standard backend for running AI workloads across diverse consumer and enterprise hardware.