Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU

Llama-modchip, a software tool, allows users to run 284B Frontier MoEs (DeepSeek-V4-Flash) and 3B NEMO models on their CPU workstation with 64GB RAM, without a GPU, by using a custom implementation of the Fractal-64 models and deepseek4 kernels. The tool provides a pre-compiled binary and a batch file to automate the process, and users can submit their benchmark scores to a leaderboard. The tool also includes documentation for optimizing performance on various processors and GPUs, and provides operator-level profiling and mathematical proofs for each model family.