怎么用单卡训练？

[xinhaojin]

如题

[arkerman]

@xinhaojin 哈喽，这个不能单卡训练吗？我只有一块GPU运行报错，好像是关于显卡的问题

[jyqi]

你好，将--nproc_per_node设为1即可进行单卡训练。

[arkerman]

@jyqi 你好帅哥，我已经按照你的提示设置了单卡，但是报了以下的错误：

2022-11-30 10:10:14.802 | ERROR | main::68 - An error has been caught in function '', process 'MainProcess' (25404), thread 'M
ainThread' (8124):
Traceback (most recent call last):

File "tools\train.py", line 68, in
main()
└ <function main at 0x00000253ED19A318>

File "tools\train.py", line 53, in main
torch.distributed.init_process_group(backend='gloo', init_method='env://')
│ │ └ <function init_process_group at 0x00000253EC862DC8>
│ └ <module 'torch.distributed' from 'D:\Anaconda3\envs\DAMO-YOLO\lib\site-packages\torch\distributed\init.py'>
└ <module 'torch' from 'D:\Anaconda3\envs\DAMO-YOLO\lib\site-packages\torch\init.py'>

File "D:\Anaconda3\envs\DAMO-YOLO\lib\site-packages\torch\distributed\distributed_c10d.py", line 421, in init_process_group
init_method, rank, world_size, timeout=timeout
│ │ │ └ datetime.timedelta(seconds=1800)
│ │ └ -1
│ └ -1
└ 'env://'

File "D:\Anaconda3\envs\DAMO-YOLO\lib\site-packages\torch\distributed\rendezvous.py", line 82, in rendezvous
raise RuntimeError("No rendezvous handler for {}://".format(result.scheme))
│ └ <property object at 0x00000253E9FE0AE8>
└ ParseResult(scheme='env', netloc='', path='', params='', query='', fragment='')

RuntimeError: No rendezvous handler for env://

这是啥原因呀？

[jyqi]

你好，看到你的backend改成了gloo，是因为在windows平台训练吗？我们在linux平台上尝试单卡训练，没有碰到过这个错误。可以排查下是否平台原因。

[arkerman]

@jyqi 好的，谢谢你的建议