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'])

【pytorch并行化常见bug】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.

    推荐阅读