Appgithub-com-generalduel-kimi-k3-in-c

Download Pinokio

Sign in Create accountJoin

Sign in Create accountJoin

kimi-k3-in-c

A 2.78-trillion-parameter Kimi K3 running inference on a single CPU in 8.24 GB of RAM. Portable C99: no BLAS, no framework, no GPU.

github.com/generalduel/kimi-k3-in-cUpdated 1mo ago

Repository details

Updated 2026-08-26T07:36:49.000Z

Indexed 2026-08-30T16:14:09.156Z

Loading community details…

Loading project activity…

Community

Sort byCreate post

Ask or share about kimi-k3-in-c…

Loading...

Pinokio

Projects using this repository

No references listed yet.

Check-ins (0)

No check-ins yet.

Recent commits

docs: add CONTRIBUTORS.md

Fareed Khan2 months ago117e9d2

build: add a CMake CI regression job and pin Python tool dependencies

Fareed Khan2 months ago68622d7

style(test): use the repo's ASCII dash convention in test_trunk.c

Fareed Khan2 months ago3144c1e

feat: add native Windows support via MinGW-w64

Fareed Khan2 months ago1f19bbe

feat: add exact ultra-low-memory full-model inference

Fareed Khan2 months agob2f5b5b

fix(core): enforce k3_matmul_mxfp4's documented preconditions

Fareed Khan2 months agoa4197ad

add NEON paths for the bf16, MXFP4 and q8 matmuls on aarch64

Fareed Khan2 months agof46c8cd

fix: macOS/Apple Silicon download script portability and Darwin pread 2 GiB ceiling

Yongsheng Gao2 months ago0fe6d0b

test: add synthetic trunk streaming regression coverage

Coreywwj2 months ago65841a1

build(cmake): support ARM64 architecture flags

Shaalan Marwan2 months ago48a813c