From the creator of Redis; run LLM locally with ds4

DwarfStar 4 (ds4), created by the founder of Redis, is a C-based inference engine that enables users to run large language models locally on high-memory machines with Metal, CUDA, or ROCm support. The tool uses asymmetric 2-bit quantization and disk-based KV caching to compress models like DeepSeek V4, GLM 5.x, and Qwen while maintaining performance, offering a CLI, local APIs, and a native agent interface. It's designed to make frontier-class models practical for personal use without relying on remote servers.