Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

`deepspeed.comm.barrier()` has different signatures/behaviour from `torch.distributed.barrier()`

See original GitHub issue

https://github.com/microsoft/DeepSpeed/blob/4abf637f96cd36d450125a38a15d21df1cf0b8db/deepspeed/comm/comm.py#L456-L458

It seems the deepspeed.comm.barrier() has different signatures/behaviour from torch.distributed.barrier(). Is this intended?

reference: torch.distributed.barrier() https://github.com/pytorch/pytorch/blob/07dd2fe6c32948e5ca0a2871e5eb31602a9684cf/torch/distributed/distributed_c10d.py#L3182

Also, is monitored_barrier() supported in deepspeed.comm? Thanks!

Issue Analytics

State:
Created 10 months ago
Comments:5 (4 by maintainers)

Top GitHub Comments

2reactions

Quentin-Anthonycommented, Dec 2, 2022

@jeffra – Sure!

I can also add monitored_barrier() support for you, @HeyangQin

1reaction

jeffracommented, Dec 2, 2022

@Quentin-Anthony can you take a look at this? I think the barrier function signature needs to support barrier(group=GroupMember.WORLD, async_op=False, device_ids=None)