GGML

ggml is a high-performance tensor library written in C optimized for Apple Silicon, offering features like 16-bit floats, integer quantization, and zero runtime memory allocations.

Machine Learning Updated 6 hours ago

Visit Website

GGML's Top Features

Written in C

16-bit float support

Integer quantization support (4-bit, 5-bit, 8-bit)

Automatic differentiation

Built-in optimization algorithms (ADAM, L-BFGS)

Optimized for Apple Silicon

Supports AVX/AVX2 intrinsics on x86 architectures

WebAssembly and WASM SIMD support

No third-party dependencies

Zero memory allocations during runtime

Frequently asked questions about GGML

ggml is a high-performance tensor library written in C that supports large models and high performance on commodity hardware.

ggml is optimized for Apple Silicon and x86 architectures and supports WebAssembly for web deployment.

ggml offers 16-bit float support, integer quantization, automatic differentiation, built-in optimization algorithms, zero memory allocations during runtime, and guided language output support.

ggml is used for applications such as short voice command detection on Raspberry Pi, running multiple model instances on Apple devices, and deploying high-efficiency models on GPUs.

Yes, ggml is open-source and available under the MIT license. The development process is open, and community contributions are encouraged.

Related projects include whisper.cpp for high-performance speech recognition and llama.cpp for efficient inference of Meta's LLaMA language model.

You can contribute to the ggml codebase or financially support the project by becoming a sponsor to contributors of llama.cpp, whisper.cpp, or ggml.

ggml.ai was founded by Georgi Gerganov with pre-seed funding from Nat Friedman and Daniel Gross.

Yes, ggml.ai is seeking full-time developers who share their vision and have contributed to related projects. Interested candidates can contact jobs@ggml.ai.

For business-related topics, you can contact ggml.ai through the information provided on their website.