完美解决windows系统raise RuntimeError(“Distributed package doesn‘t have NCCL “

Posted 2021-05-24 ai52learn

在训练时出现如下问题： 

File "C:\\Users\\urser\\anaconda3\\lib\\site-packages\\torch\\distributed\\distributed_c10d.py", line 597, in _new_process_group_helper
    raise RuntimeError("Distributed package doesn't have NCCL "
RuntimeError: Distributed package doesn't have NCCL built in

从文字上来看，错误提示很明显了，没有NCCL

而windows不支持NCCL backend.

我们看下官方文档：

As of PyTorch v1.8, Windows supports all collective communications backend but NCCL, If the init_method argument of init_process_group() points to a file it must adhere to the following schema:

而要解决这个问题也很简单，不使用NCCL backend.就可以了。

只需要一行代码就可以解决问题。

获取解决方案：https://ai.52learn.online/11955