Running Kimi K3 on a M1 Max

Deltafin is a research project that successfully runs Kimi K3, a 2.8 trillion parameter Mixture-of-Experts model, on an Apple Silicon M1 Max laptop with 64GB of RAM. The system operates at approximately 16 seconds per token in full mode (with 1.7TB of local storage) or 3+ minutes per token in streaming mode, making it slow but functional and reproducible. The project includes installation tools and flexible deployment options allowing users to start with streaming mode and upgrade to faster local storage later.