Pytorch“NCCL 错误”：未处理的系统错误，NCCL 版本 2.4.8”

Question

我使用pytorch分布式训练我的模型。我有两个节点和每个节点两个gpu，我为一个节点运行代码：

python train_net.py  --config-file configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_coco.yaml  --num-gpu 2  --num-machines 2 --machine-rank 0 --dist-url tcp://192.168.**.***:8000

和另一个：

python train_net.py  --config-file configs/InstanceSegmentation/pointrend_rcnn_R_50_FPN_1x_coco.yaml  --num-gpu 2  --num-machines 2 --machine-rank 1 --dist-url tcp://192.168.**.***:8000

但是另一个有 RuntimeError 问题

global_rank 3 machine_rank 1 num_gpus_per_machine 2 local_rank 1
global_rank 2 machine_rank 1 num_gpus_per_machine 2 local_rank 0
Traceback (most recent call last):
  File "train_net.py", line 109, in <module>
    args=(args,),
  File "/root/detectron2_repo/detectron2/engine/launch.py", line 49, in launch
    daemon=False,
  File "/root/anaconda3/envs/PointRend/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 171, in spawn
    while not spawn_context.join():
  File "/root/anaconda3/envs/PointRend/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 118, in join
    raise Exception(msg)
Exception:

-- Process 0 terminated with the following error:
Traceback (most recent call last):
  File "/root/anaconda3/envs/PointRend/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 19, in _wrap
    fn(i, *args)
  File "/root/detectron2_repo/detectron2/engine/launch.py", line 72, in _distributed_worker
    comm.synchronize()
  File "/root/detectron2_repo/detectron2/utils/comm.py", line 79, in synchronize
    dist.barrier()
  File "/root/anaconda3/envs/PointRend/lib/python3.6/site-packages/torch/distributed/distributed_c10d.py", line 1489, in barrier
    work = _default_pg.barrier()
RuntimeError: NCCL error in: /pytorch/torch/lib/c10d/ProcessGroupNCCL.cpp:410, unhandled system error, NCCL version 2.4.8

如果我把mask-rank=1改成mask-rank=0，那么就不会报错，但是不能分布式训练，有谁知道为什么会出现这个错误？

Answer 1

许多因素都可能导致此问题，例如参见1、2。添加行

import os
os.environ["NCCL_DEBUG"] = "INFO"

您的脚本将记录导致错误的更具体的调试信息，为您提供更有用的错误消息给谷歌。