Pytorch Pytork:DataParallel can'；t在总批量大小相同的情况下减少模型推理时间
当我使用Pytorch Pytork:DataParallel can'；t在总批量大小相同的情况下减少模型推理时间,pytorch,Pytorch,当我使用torch.nn.DataParallel（）实现数据并行计算时，我发现对于相同的总量批量大小，并行模型的推理时间与串行模型相比没有显著减少 如下图所示 model = SNModel() criterion = nn.CrossEntropyLoss() if parallel_enable: model = nn.DataParallel(model, device_ids=gpu_ids) # gpu_ids = [0,1,2,3] total 4 gpus model.t
torch.nn.DataParallel（）model = SNModel()
criterion = nn.CrossEntropyLoss()
if parallel_enable:
model = nn.DataParallel(model, device_ids=gpu_ids) # gpu_ids = [0,1,2,3] total 4 gpus
model.to(args.device)
criterion.to(args.device)
torch.cuda.synchronize()
st = time.time()
outputs, loss = model(images, path, labels, criterion, gpu_nums)
torch.cuda.synchronize()
total_tm += time.time() - st
100 iter model time 23s (around)
100 iter model time 103s (around)
103-23*4=11s