GitHub – antirez/ds4: DeepSeek 4 Flash and PRO local inference engine for Metal, CUDA and ROCm · GitHub
Local Deepseek-v4-Flash, recommend 128GB memory
GitHub – JustVugg/colibri: Run GLM-5.2 (744B MoE) on a 25GB-RAM consumer machine — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦 · GitHub
Run GLM 5.2 with only 24GB unified memory. Downside is 0.1 – 1 token per second, so super slow. Could be useful for scheduled overnight reasoning tasks.