TT-Metal is a comprehensive ML framework providing TT-NN, a neural network operator library written in C++ and Python, alongside TT-Metalium, a low-level kernel programming model. It is designed to optimize and run large language models (Llama, DeepSeek), image generation (Stable Diffusion), and video generation on Tenstorrent hardware accelerators. The project includes performance benchmarking tools for measuring token throughput, latency, and parallelization across multiple devices.