Help Needed from vLLM team on profiling pytorch cuda memory

Answered here.