#c_lang
DwarfStar is a fast, narrow AI engine built mainly for DeepSeek V4 Flash, with support for GLM 5.2 and, on very large-memory machines, DeepSeek V4 PRO. It runs on Mac Metal, NVIDIA CUDA, and ROCm, and can use SSD streaming or distributed setups when models do not fully fit in RAM. The benefit is that you can run powerful local models on your own hardware, use them for chat, coding, or server workloads, and keep control of speed, memory, and cost.
https://github.com/antirez/ds4
Post #15903
916