AI NewsOpen sourceAnnouncement
llama.cpp 1.0 commits to a stable C API across 150 model types
The ggml team tagged llama.cpp and ggml v1.0.0 and promised to bump the major version on every breaking change.
Image: GitHub
Why it mattersTeams pinning llama.cpp to a version now get a public contract. A minor upgrade will not break the C API. Breaking changes only come with a major bump.
A team running local inference on llama.cpp had no easy way to tell when an upgrade would break its code. The ggml team closed that gap.
The authors tagged llama.cpp v1.0.0 and ggml v1.0.0 on Hugging Face, and committed to semantic versioning for both. The authors are Daniel Bevenius, Georgi Gerganov and the ggml-org team. Gerganov started the project about three and a half years ago, with the first commit in August 2023.
What v1.0.0 changes
The headline is a rule, not a feature. The authors write: "Moving forward, both llama.cpp and ggml will reflect API and ABI incompatibilities with major-version bumps. As adoption grows, we will be stricter about making those boundaries visible to downstream consumers."
The scope of each bump is spelled out. A major bump marks a breaking change to the public C API in ./include. A minor bump covers a backward-compatible feature, a backend update, a new model, or an API addition. A patch is a bug fix with no API change. Everything outside the public header folders is still internal.
What the library covers
llama.cpp runs multi-modal large language models across six hardware families named in the post: NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. The authors say the library supports more than 150 model architecture variants and every mainstream speculative decoding method.
Builds go out as pre-release files on Linux, macOS, Windows, Android and iOS. They are also on Docker, Debian, Homebrew, conda-forge, MacPorts and Winget. There is an XCFramework for Apple platforms. The authors also say Hugging Face Buckets and HF Jobs now scale the project's continuous integration.
The authors give one reason for the whole move. "Every new integration means more developers depending on our library interfaces," they write. So "a flexible but stable API is paramount."
The practical change for a team running llama.cpp in production is small but useful. A minor or patch upgrade now comes with a public promise that the C API did not change. The team can patch for a new model or a backend fix without a code change. A planned major upgrade reads as the day the inference server needs attention.
Source
llama.cpp v1.0.0: A stronger foundation for the local AI ecosystem, Hugging Face blog by Daniel Bevenius, Georgi Gerganov and the ggml-org team.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
