microsoft/DeepSpeed

AttributeError: module 'torch.distributed' has no attribute 'ReduceOp'

SreehariSankar opened this issue · 4 comments

Cannot import DeepSpeed. Above mentioned error obtained when I tried both "Import deepspeed" and "ds_report".

Mac OSX 12.3
Conda 22.9.0
Python 3.8.16
pytorch 1.12.1
pytorch-ignite 0.4.10
pytorch-lightning 1.9.0
transformers 4.21.3

installed using "pip install deepspeed"

please help

@SreehariSankar, can you please share a stack trace of the failure during deepspeed install? We are not aware of deepspeed working on OSX because of the current CUDA dependency. This could be the reason, but would like to confirm. Thanks!

DS installed properly. The above error was returned
Screen Shot 2023-02-16 at 1 05 55 PM

Just to clarify, I think you are right, because I tried this on a CUDA-installed CentOS machine and everything was OK (Same package list and version).
So i am inclined to believe that this is a MacOS issue.

@SreehariSankar, do you think this issue should remain opened? Thanks!

Closed! Thanks for the clarification on DeepSpeed and MacOS