# ashhart/TensorFold

TensorFold: High\-Performance LLM Inference Engine for Metal, CUDA, and Vulkan

- HackerByte page: <https://hackerbyte.net/en/projects/tensorfold-high-performance-llm-inference-engine-for-metal-cuda-and-vulkan-1274723381>
- GitHub: <https://github.com/ashhart/TensorFold>
- Programming language: Zig

LLM Inference Engine for Metal, CUDA and Vulkan.

## AI-generated summary

TensorFold is a high\-performance LLM inference engine written in Zig, designed to run on Apple Silicon \(Metal\) and NVIDIA GPUs \(CUDA, Vulkan\). It allows serving language models directly from a Zig binary without requiring Python or MLX installations, offering a streamlined experience for developers.

## Key points

- Supports LLM inference on Apple Silicon \(Metal\) and NVIDIA GPUs \(CUDA, Vulkan\).
- Written in Zig for efficient, standalone binary deployment.
- Eliminates the need for Python or MLX dependencies for model serving.
- Offers a quick start guide with various models like Qwen3.5\-2B on Mac and Linux.
- Includes advanced features like Living Weights for dynamic model learning and a desktop workspace for a comprehensive development environment.