[trimul] Calculate Cauchy distribution on GPU #66
Conversation
This file contains hidden or 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
While running eval.py locally, I observed super long iteration times for any Cauchy distributions. Profiles show that most of the time is spent on the CPU generating Cauchy distributions. For example, on a H100, I saw a 100-iteration benchmarking run for a BS=2, dim=256, seqlen=128, cauchy distribution take 13 seconds without this change, and <1s after this change. My suspicion is that the previous behavior could lead to some submissions timing out if they have more variation than average.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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.
While running eval.py locally, I observed super long iteration times for any Cauchy distributions. Profiles show that most of the time is spent on the CPU generating Cauchy distributions.
For example, on a H100, I saw a 100-iteration benchmarking run for a BS=2, dim=256, seqlen=128, cauchy distribution take 13 seconds without this change, and ~0.3s after this change. For such a distribution (i.e. generating [2, 128, 128, 256]), torch profiler shows the cauchy data generation is taking ~113ms
My suspicion is that the previous behavior could lead to some submissions timing out if they have high variation, such that they trigger a full 100 iterations on some large cauchy-distribution test cases