LLM-d

Prefill-decode disaggregation, now supported by vLLM, SGLang, LLM-d, NVIDIA Dynamo, and TensorRT-LLM, separates LLM inference into compute-bound prefill and memory-bound decode phases on different nod…