The llama.cpp project has released libraries v1.0.0, according to a post on the Hugging Face blog. The release sets a versioning policy for the llama.cpp and ggml libraries. Under that policy, API and ABI incompatibilities will be marked with major-version bumps.

The post says llama.cpp can run multi-modal LLMs on hardware ranging from consumer machines to enterprise servers. It credits quantization, optimized backends, and a portable C/C++ implementation for this. The project began 3.5 years ago, and the post states that many downstream applications already depend on it.

The ggml engine runs on dedicated backends. Each backend can be compiled as a dynamic library and loaded at runtime. The post says llama.cpp supports more than 150 architecture variants and all mainstream speculative decoding methods.

The post says the project's continuous integration infrastructure now uses HF Buckets and HF Jobs. It also says self-hosted runners are being added across major hardware platforms, with hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel.

The post states that the team uses LLMs in its development cycle, and that maintaining code quality and architectural control remain its top priorities