Dev News Daily ENDE

llama.cpp and ggml reach v1.0.0 and tie major versions to breaks in the public C API

The ggml-org team has released version 1.0.0 of both llama.cpp, the C/C++ library for running language models locally, and ggml, the tensor library underneath it. The announcement on the Hugging Face blog, signed by Daniel Bevenius, Georgi Gerganov and Merve, presents the number less as a feature milestone than as a commitment to the developers who build on the two libraries.

From now on, both projects will mark incompatibilities in their API and ABI with a major-version bump. The rules apply to every ggml-org project that uses semantic versioning: a breaking change to the public C API in ./include raises the major version; a backward-compatible feature, backend update, model support or API addition raises the minor version; a bug fix with no API change raises the patch number. The public headers are llama.cpp/include/ for libllama and ggml/include/ for libggml. The team writes that as adoption grows it will be stricter about making those boundaries visible to downstream users.

The post gives the project's age as three and a half years and says many downstream applications already depend on it. It lists the current distribution channels: pre-release artifacts for Linux, macOS, Windows, Android and iOS, plus Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for iOS and macOS.

On development practice, the team says it uses LLMs in its development cycle but keeps control of the architecture and of implementation details as top priorities. CI capacity has been expanded with Hugging Face Buckets and Jobs, and hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel adds self-hosted runners across the main hardware platforms.

ggml reaches that hardware through dedicated backends, each of which can be built as a dynamic library and loaded at runtime. The announcement states that llama.cpp supports more than 150 architecture variants and all mainstream speculative decoding methods.

llama.cpp and ggml reach v1.0.0 and tie major versions to breaks in the public C API
llama.cpp and ggml reach v1.0.0 and tie major versions to breaks in the public C API — Dev News Daily

Why it matters

Until now a consumer of libllama had no version number that said whether an update would break the build. With the headers in ./include named as the contract, packagers and application authors can pin a major version and take minor and patch updates, which is the precondition for the distribution channels the post lists.