Its repository description says it runs inference for a 2.78-trillion-parameter Kimi K3 model on a single CPU in 8.24 GB of RAM, with no BLAS dependency, framework, or GPU requirement. The repository includes source, benchmarks, tests, examples, and build files; the supplied source snapshot also shows 8.7k stars and 1.4k forks.
Raw SignalSep 28, 20261 min read1 source
kimi-k3-in-c: A 2.78-trillion-parameter Kimi K3 running inference on a single CPU in 8.24 GB of RAM
In one minute
The public GitHub repository `FareedKhan-dev/kimi-k3-in-c` presents an Apache-2.0, portable C99 inference engine for Kimi K3.