Chris Van Pelt (CVP)
Chris Van Pelt (CVP)
It actually works with 3.0.0, you just need to pass in the low level api to start. ```python image = "nginx" dir = "/app" cwd = os.getcwd() container = client.api.create_container(...
Not yet but we're actively working on improvements to our sweep infrastructure. cc @dannygoldstein
Hey @yaroslavvb I could imagine a `disable` keyword arg to `wandb.init`. Until then you can do the following to achieve what you want: ```python import os import wandb if rank!=0:...
This is a great writeup! We'll discuss internally.
@tshrjn This likely won't be addressed this quarter. What's the most important part of this discussion for your workflows?
Hey @kwanUm, @joshlk, @kanybekasanbekov apologies for the delay in response here. wandb does support distributed training natively. This thread is about more complex distributed cases. When training in a distributed...
Hey @sumanthratna thanks for this. We'll look into the root cause of these new failing tests.
> Any reason why the new PR was necessary? It seems to be the exact same code as the original with the comment deleted and the linter is still complaining...
An important thing to consider here is how a user easily resumes from a checkpoint. Making them specify resume in init is not ideal. Making them specify WANDB_RESUME_FROM_CHECKPOINT feels rather...
> @vanpelt What about a `latest=True` argument to `log_artifact`? By default, we always mark things as `latest` unless the user explicitly tells us not to. Nice, yep that's better than...