GGML

GGML Overview

GGML is an advanced tensor library designed for machine learning. It enables large models and high performance on standard hardware. Written in C, GGML supports 16-bit float and integer quantization (4-bit, 5-bit, 8-bit). It features automatic differentiation, built-in optimization algorithms, and is optimized for Apple Silicon. GGML also supports x86 architectures and web via WebAssembly and WASM SIMD. It has no third-party dependencies and allows zero memory allocations during runtime. GGML is used by llama.cpp and whisper.cpp, and is developed by ggml.ai.

GGML Highlights

• Supports large models and high performance on commodity hardware
• Optimized for Apple Silicon and supports x86 architectures
• Features automatic differentiation and built-in optimization algorithms