KEMBAR78
Lazy init order in set device, should not be called in getDevCount by csarofeen · Pull Request #4918 · pytorch/pytorch · GitHub
Skip to content

Conversation

@csarofeen
Copy link
Contributor

No description provided.

@csarofeen
Copy link
Contributor Author

#4903

Copy link
Contributor

@apaszke apaszke left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you please revert the pybind submodule update?

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

@csarofeen
Copy link
Contributor Author

The biggest issue seems to be _lazy_init as it is allocating a good chunk of memory that can end up on the wrong device fairly easily the way things are now. Calling device count in pytorch was allocating about 1.5GB on my first device. Device count in pure CUDA was creating ~10MB on each device.

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

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants