ggml-org/ggml

Health v2 · maintenance only

Adoption not part of the score

15236 stars · 1792 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

ggml is a plain C/C++ tensor library for machine learning with no dependencies, designed to be simple, portable, and efficient. It supports SIMD-optimized kernels across many CPU architectures plus CPU, GPU, NPU, and browser backends, with integer quantization and zero runtime memory allocations.

Use cases

• run LLM inference on CPU at the edge
• embed a tensor engine in a C/C++ app with no dependencies
• quantize models to 2-8 bit integers for low-memory inference
• run ML inference in the browser via WebAssembly
• build custom ML inference engines like llama.cpp or whisper.cpp

When to choose

• you need dependency-free, portable ML inference in C/C++
• you target edge devices, mobile, or WebAssembly
• you want fine-grained control over memory and quantization

When to avoid

• you need automatic differentiation for training large models
• you prefer high-level frameworks like PyTorch or TensorFlow
• you need a broad ecosystem of pretrained model loaders and tools

Facets

library · maturity stable

machine-learning llm-inference gpu-computing wasm machine-learning deep-learning large-language-models gpu-computing embedded-systems cross-platform cpp c wasm embedded tensor-library quantization inference no-dependencies simd gguf gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="ggml-org/ggml")