The team behind llama.cpp, an open-source tool for running AI models on local machines, released version 1.0.0 of its libraries. It said so in a post on the Hugging Face blog dated 13 November. The project is about 3.5 years old. The team said both llama.cpp and ggml, the engine underneath it, will now raise the major version number whenever a change breaks API or ABI compatibility. Those terms cover how other programs call the code and how compiled builds link to it.
The post says llama.cpp now supports more than 150 model architecture variants. It also says the project covers all mainstream speculative decoding methods. The team said it is adding self-hosted test machines, with hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. It also uses Hugging Face tools to run checks on every change to its main branches. The text does not say what changed in 1.0.0 itself. It does not say whether any existing interface breaks, and it does not set out the versioning rules in full.
Many downstream apps depend on llama.cpp, the team said. For people who build or run software on it, a major version bump now marks a change that can break existing code. That gives teams a clear signal to check before they upgrade.