Tools
kimi-k3-in-c Runs a 2.78-Trillion-Parameter Model on One CPU in 8.24 GB of RAM, and Hits 6,131 Stars in 19 Days
FareedKhan-dev/kimi-k3-in-c is a 176 KB portable C99 engine with no BLAS, no framework and no GPU that streams a 1.56 TB Kimi K3 checkpoint off disk, with a measured peak RSS of 8.24 GB. The README's own table shows 26.5 s per token on an 8 GB laptop falling to 5.6 s at 128 GB+, with byte-identical output at every memory size, so more RAM buys only speed. It has taken 6,131 stars since its 2026-08-01 creation, roughly 323 a day, and v1.0.0 claims about 8x lighter math per token than the first cut.
Source
↳ Follow the thread