llama.cpp and ggml reach v1.0.0 with a promise to flag breaking API changes

The local AI inference project now follows semantic versioning, so a major-version bump will mark any incompatible change to its public C interface.

CodePublished
Image published by Hugging Face with “llama.cpp v1.0.0 : A stronger foundation for the local AI ecosystem”
Image: Hugging Face · Source
What happened
The llama.cpp team at Hugging Face released version 1.0.0 of both the llama.cpp and ggml libraries on 13 November.
What’s new
Both projects now commit to semantic versioning, with a major-version bump reserved for breaking changes to the public C API in their include headers.
Why it matters
Applications built on libllama or libggml can now tell from a version number alone whether an update may break their integration.

The llama.cpp team at Hugging Face released version 1.0.0 of two libraries on 13 November: libllama, which performs LLM inference in C/C++, and libggml, the tensor library underneath it. The project is three and a half years old and runs multi-modal models on hardware that ranges from consumer machines to enterprise servers.

The main change is a versioning commitment. In the announcement, the team says a growing number of downstream applications rely on its library interfaces, which makes a stable API essential. Under the new rules, which cover all ggml-org projects, a major version marks a breaking change to the public C API in the include headers. A minor version covers backward-compatible additions, backend updates and new model support. A patch version is for bug fixes that leave the API untouched. The team says it will be stricter about signalling API and ABI incompatibilities to downstream users.

The libraries are distributed as pre-release artifacts for Linux, macOS, Windows, Android and iOS, and through Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for Apple platforms.

According to the announcement, llama.cpp supports more than 150 architecture variants. Its testing infrastructure runs on Hugging Face's HF Buckets and HF Jobs, with self-hosted runners on hardware from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel.