pipenetwork/Kimi-K2.7-Code-MLX-4bit-hiprec
Text Generation • 1T • Updated • 829 • 1
MLX build of Kimi-K2.7-Code. Base is natively 4-bit (int4 experts + bf16 rest); this keeps experts at 4-bit and lifts non-expert layers to 6-bit.
Note experts@3-bit, attention/router/embeds/shared/dense@6-bit · fits 512GB · smoke-tested