pytorch MAC 측정 thop vs ptflops
결론 : ptflops가 더 좋음 + thop는 모듈의 옵션을 반영하지 않음 + thop는 Attention 모듈 지원하지 않음 1. m = nn.Conv2d(1,2,(3,3),(1,1),bias=False) m.eval() input = torch.rand(1,1,10,10) macs_thop, params_thop = profile(m, inputs=(input,)) print("thop : MACS : {} | Param : {}".format(macs_thop,params_thop)) # https://github.com/sovrasov/flops-counter.pytorch macs_ptflos, params_ptflops = get_model_complexity_info(m, (1,10,10)..
2024. 2. 8.