utilization.md

GPU Utilization Profiling Usage:

from utilization import gpuUtilization, UtilizationContext, UtilizationTrainContext

...
with UtilizationContext(args.get_util) as context:

    for epoch in epoch:
        for sample in samples:
            with UtilizationTrainContext(context):
                ...
                # this is a pass
                model.train()
                forward()
                backward()
                ...

Recording result:

exp_recorder.record("utilization", context.getAvg())

Also, need to increase iteration number for get speed a little bit (5->~18)