DwarfStar

DwarfStar is a narrow, self-contained native local LLM inference engine for running supported DeepSeek V4 Flash/PRO and GLM 5.2 GGUF model layouts on Metal, CUDA, and ROCm hardware. It includes model loading, prompt rendering, tool calls, KV state, an HTTP server, a CLI, and a coding agent, and is not intended as a general GGUF runner.