Setting OMP_NUM_THREADS

Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 

1.  
   (pytorch1.3-cuda10.2) [huanghaiyang@dgx02 semantic-segmentation-main]$ python -m runx.runx scripts/eval_cityscapes.yml -i
2.  
   *****************************************
3.  
   Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
4.  
   *****************************************
5.  
   None
6.  
   None
7.  
   NoneNone
8.  
    
9.  
   None
10.  
    None
11.  
    Global Rank: 5 Local Rank: 5
12.  
    Global Rank: 1 Local Rank: 1
13.  
    Global Rank: 2 Local Rank: 2
14.  
    Global Rank: 3 Local Rank: 3
15.  
    Global Rank: 0 Local Rank: 0
16.  
    Global Rank: 6 Local Rank: 6
17.  
    None
18.  
    Global Rank: 7 Local Rank: 7
19.  
    None
20.  
    Global Rank: 4 Local Rank: 4

跑代码的时候出现上述问题，这是由于pytouch分布式训练的问题。别人的代码设置了多个GPU并行，然而你跑的时候只用了一个或者两个，这个参数需要指定

python -m torch.distributed.launch --nproc_per_node=1 --master_port 88888 train.py

--nproc_per_node=1 这个1表示你实际的GPU数量，

--master_port 88888 这个表示端口，一般不用设置，或者随便设置一个数字就行。当出现

runtimeerror: address already in use

这时候加--master_port 12345 就行

这篇好文章是转载于：学新通技术网