DwarfStar
Native C inference engine for running a few large open models locally
What it is
DwarfStar is a small, self-contained inference engine written in C that runs a short list of large open-weight models, mainly DeepSeek V4 Flash and PRO plus GLM and Qwen variants, on Metal, CUDA and ROCm. It ships its own GGUF files, an interactive CLI, an HTTP server for multiple sessions, and a built-in coding agent, with SSD streaming, tensor parallelism across RDMA-linked Macs, and multi-GPU support. People with high-memory Macs, DGX Spark, Strix Halo boxes or CUDA card rigs use it to run frontier-class open models on hardware they own.