improve device picking #113
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
matsen
approved these changes
Feb 12, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Pick the device as before, except that when picking the CUDA GPU, if there is no difference between gpu utilization, check whether there is a significant difference in allocated memory, and then check which gpu has the fewest running processes.
Also, to immediately register the result of pick_device with nvidia-smi, if pick_device returns a CUDA GPU, it also immediately sends a tensor to that chosen device.
This removes the need to pick devices twice in
trainfunctions, with the second pick_device called after a random waiting period. I think we still need the random waiting period, but it can be shorter, and it can be immediately before the first call to pick_device. Currently, the waiting period is rarely sufficient to allow any usage to show up in nvidia-smi from other simultaneous processes. (These changes will be implemented in a separate dns-experiments PR)