Skip to content

DistributedLossWrapper enhancements#702

Merged
KevinMusgrave merged 11 commits intoKevinMusgrave:devfrom
elisim:master
Jul 23, 2024
Merged

DistributedLossWrapper enhancements#702
KevinMusgrave merged 11 commits intoKevinMusgrave:devfrom
elisim:master

Conversation

@elisim
Copy link
Copy Markdown
Contributor

@elisim elisim commented Jul 17, 2024

fixes #701

I've also added a warning when DistributedLossWrapper is being used in a non-distributed setting. Currently, an error is raised because of:

world_size = torch.distributed.get_world_size()

@elisim
Copy link
Copy Markdown
Contributor Author

elisim commented Jul 18, 2024

@KevinMusgrave It appears that the test failures are unrelated to the PR.

A module that was compiled using NumPy 1.x cannot be run in
NumPy 2.0.0 as it may crash. To support both 1.x and 2.x
versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind[11](https://github.com/KevinMusgrave/pytorch-metric-learning/actions/runs/9976990884/job/27577893340?pr=702#step:5:12)>=2.12'.

@KevinMusgrave
Copy link
Copy Markdown
Owner

@elisim Thanks, I'll look into it

@elisim
Copy link
Copy Markdown
Contributor Author

elisim commented Jul 22, 2024

ping @KevinMusgrave :)

@KevinMusgrave KevinMusgrave changed the base branch from master to dev July 23, 2024 14:20
@KevinMusgrave KevinMusgrave merged commit 0db2483 into KevinMusgrave:dev Jul 23, 2024
@KevinMusgrave
Copy link
Copy Markdown
Owner

Thanks @elisim !

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

arguments names are different in forward in when using the DDP wrapper

2 participants