多GPU训练的模型,转换加载到单GPU测试或预训练
报错:RuntimeError: Attempting to deserialize object on CUDA device 3 but torch.cuda.device_count() is 1.
加载的模型是3个GPU进行训练的,而你的计算机只有1个GPU
state_dict = model.state_dict()
weights_path = torch.load(opt.weights_path)['model_state_dict']
model.load_state_dict(weights_path)
解决方案:
torch.load()参数中,写入参数map_location=‘cuda:0’
state_dict = model.state_dict()
weights_path = torch.load(opt.weights_path, map_location='cuda:0')['model_state_dict']
model.load_state_dict(weights_path)
如果GPU是多块到2块:就把map_location改为:map_location={‘cuda:1’: ‘cuda:0’}。其余情况依次类推。