Understanding Communication Bottlenecks in Multi-node LLM Inference

Updated: