
kimi-k3-in-c
Run a 2.78-trillion parameter MoE model on 8GB RAM using pure C.
avx2c99cpu-inferencedeep-learning
Kimi-k3-in-c is a lightweight, portable C99 inference engine that allows consumer hardware to run massive Mixture of Experts AI models without needing GPUs, heavy frameworks, or complex clusters. By streaming sleeping experts directly from disk, it reduces a 1.56 TB model footprint down to just 8.24 GB of RAM on a single CPU. It is designed for developers, researchers, and AI enthusiasts who want to experiment with trillion-parameter language models locally on ordinary machines.