I would like to use DeepEP when running DeepSeek-V3 in SGLang, an LLM serving framework. DeepEP requires IBGDA (InfiniBand GPUDirect Async), and I followed the setup instructions by modifying the NVIDIA driver. Is it possible to use the IBGDA feature in instant clusters?
https://github.com/deepseek-ai/DeepEP/blob/main/third-party/README.md...