In the context of large language model inference (vLLM, SGLang, TensorRT-LLM, etc.), here's the distinction:

## Raw