GLM 5.2 4bit Achieves 35 T/s Generation on DGX Station with DwarfStar

GLM 5.2 4bit (500GB total weights) runs at 35 t/s generation, 2000 t/s prefill on DGX Station with DwarfStar mixed RAM/VRAM inference, soon likely to reach 3k t/s.

Sources

@antirez

Sat, 22 Aug 2026 14:20:27 GMT