问题怎么定位的呢，把报错信息一股脑甩给chatgpt，问他什么意思，他就反馈说“在标准的Python库中，div() 函数不接受 rounding_mode 参数。”（虽然这个问题也不难，但是改偷的懒还是要偷）

问题再现

Traceback (most recent call last):
  File "first_stage.py", line 447, in <module>
    main()
  File "first_stage.py", line 127, in main
    return main_worker(args, logger)
  File "first_stage.py", line 132, in main_worker
    model = build_LEModel(args)
  File "/home/algroup/chenao/SPML-LAC-master/lib/models/LEModel.py", line 135, in build_LEModel
    backbone = build_backbone(args)
  File "/home/algroup/chenao/SPML-LAC-master/lib/models/backbone.py", line 179, in build_position_encoding
    position_embedding = PositionEmbeddingSine(N_steps, normalize=True, maxH=args.img_size // downsample_ratio, maxW=args.img_size // downsample_ratio)
  File "/home/algroup/chenao/SPML-LAC-master/lib/models/backbone.py", line 146, in __init__
    pe = self._gen_pos_buffer()
  File "/home/algroup/chenao/SPML-LAC-master/lib/models/backbone.py", line 159, in _gen_pos_buffer
    dim_t = self.temperature ** (2 * (torch.div(dim_t, 2, rounding_mode='floor')) / self.num_pos_feats)
TypeError: div() got an unexpected keyword argument 'rounding_mode'

原因是torch=1.7.0不支持该参数，要到1.8以后才行；
解决方案如下：两种方法选一种

1. 升级版本

安装的是pytorch-gpu版本，要和显卡对应的，我懒得折腾；

2. 在外面套一层torch.floor

torch.floor(torch.div(...))


结果是一样的；

参考

官网介绍torch.div()
TypeError: div() got an unexpected keyword argument ‘rounding_mode’ 一位老哥的博客