netrunner(@plotarmordev):dgpp, a C++/CUDA engine for DGX Spark, developed a system where each rank keeps its model slice resident on the GPU and boots from a per rank image cache in 15–30s, depending on the model. I just looked at adopting it, but our EXL3/DFlash2 and deepseek recipes aren't supported so far. Watching for compatible support...

netrunner

@plotarmordev

Cofounder & engineer. Running local AI, building things with it, and sharing what I learn.

加入 April 2022

675 正在關注 7.1K 粉絲

netrunner@plotarmordev

2026.09.26 14:12

dgpp, a C++/CUDA engine for DGX Spark, developed a system where each rank keeps its model slice resident on the GPU and boots from a per rank image cache in 15–30s, depending on the model. I just looked at adopting it, but our EXL3/DFlash2 and deepseek recipes aren't supported so far. Watching for compatible support...