ggml-org/ggml
- Stars
- 15,893
- Language
- C++
- License
- MIT
- Daily
- rank 8 · +53 stars gained
Repository metrics were observed on this edition date and may have changed.
Key details A machine learning tensor library written in C/C++. It provides multiple computation backends and low-bit quantization formats. The README highlights minimal dependencies and a design without runtime memory allocation. Use cases Build from source, then begin integrating tensor operations with the commented matrix multiplication example. Suggested use (inference): When developing custom machine learning execution code, it can be considered as a basis for comparing the required operations, quantization, and backend support. Costs and limitations The library is MIT-licensed, and the supplied materials specify no mandatory paid API or subscription. Build and execution equipment and development and operating costs for the selected backend are separate. Required resource capacity and specific cost terms need to be checked. Adoption considerations Knowledge of C/C++ and CMake builds is required. The materials alone cannot establish whether a model can run on specific equipment or what performance it would achieve. Check the APIs and operations removed in v1.0.0 and the RPC protocol changes. Contributions involving core changes are directed to llama.cpp. Why it stands out It was observed at No. 8 on daily Trending, with an increase of 53. Version v1.0.0, listed on 2026-11-13, includes cleanup of older public APIs, FP8 support, and changes to multiple backends, providing a reason to review compatibility with existing integrations.
View on GitHub