antirez / ds4 Local inference engine

Runs DeepSeek 4 on the GPU you already own.

ds4 is a local inference engine for DeepSeek 4 Flash and DeepSeek 4 PRO, built for Metal, CUDA and ROCm. Same engine, three backends, no service in the middle.

●DeepSeek 4 Flash ●DeepSeek 4 PRO

Metal

Apple GPUs

CUDA

NVIDIA GPUs

ROCm

AMD GPUs

github.com/antirez/ds4 Design study by Goose. Not affiliated with the project.