单GPU训练问题

[liwsng]

您好，非常感谢您分享的工作，我尝试在kinetics数据集上进行测试，并将GPU数量设置为1，但是在slowfast/datasets/utils.py->sampler = DistributedSampler(dataset) if cfg.NUM_GPUS >= 1 else None处报错没有初始化RuntimeError: Default process group has not been initialized, please make sure to call init_process_group.
于是我在该函数前面添加初始化功能：
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '12355'
dist.init_process_group(backend='nccl', rank=0, world_size=1)
结果在dist.init_process_group处报错初始化了两次RuntimeError: trying to initialize the default process group twice!
对于该问题没什么头绪，想寻求一些建议。

[Andy1621]

Please check #11