OpenWeightsTerminal
Pairs

GLM 5.3

2× DGX Station GB300 · NVFP4 · SGLang

Decode · tok/s
81.5
Prefill
—
tok/100W
5.8
I ran fused GLM-5.3 NVFP4 on two DGX Station GB300s, SGLang: 81.5 tok/s single stream.

First-person 2× GB300. Rank is the quoted 81.5 single stream. Not the 8× Spark vLLM print.

Officialclaimed
Open pair