Tag
User seeks community advice on reducing VRAM usage and freeing OS RAM when serving DeepSeek-V4-Flash-0731 on two DGX Spark machines with vLLM, sharing detailed configuration and memory measurements.
This paper presents SLAI T-Rex, a full-parameter post-training optimization framework for trillion-parameter MoE models on Ascend NPU SuperPOD, achieving 34.22% MFU and outperforming GPT-5.4-Mini on Operations Research tasks by 3.98 percentage points.