关于pytorch:pytorch并行化常见bug

state_dict = torch.load(opts.checkpoint)try:    trainer.net.load_state_dict(state_dict['net_param'])except Exception:    trainer.net = torch.nn.DataParallel(trainer.net)    trainer.net.load_state_dict(state_dict['net_param'])

This is for dealing a checkpoint trained in parallel.

try:     out = trainer.net.forward()except:    out = trainer.net.module.forward()

Simialarly, the net need to be transformed to module for being compatible with paralelly trained model.