llmperfscope

Whole-lifecycle profiler for LLM inference engines: load, warmup, prefill, decode, serving and teardown; speed, memory and energy. Works with any engine, needs no code changes.