Run sampler as a method if available#16888
Merged
larryliu0820 merged 1 commit intomainfrom Jan 27, 2026
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16888
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 21 PendingAs of commit ddb0dce with merge base e4060ee ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
99fd591 to
69b833e
Compare
Gasoonjia
approved these changes
Jan 27, 2026
Contributor
Gasoonjia
left a comment
There was a problem hiding this comment.
Perf is crazy!! THanks for the great work!
And we now is still load generate token from gpu to cpu one by one right iiuc. if so the perf should be boost again after we use batch transfer!
69b833e to
a1ae2bc
Compare
Closed
a476fa8 to
800721a
Compare
With this PR: huggingface/optimum-executorch#207 we are adding a new method "sampler" to ASR models, alongside with "encoder" and "text_decoder". The flow becomes: if temperature is 0 and sampler method is available, run that method. Otherwise still go with the old path. This change should largely improve the performance on CUDA since we don't have to copy logits from device to CPU for sampling purpose. Benchmark result:
800721a to
ddb0dce
Compare
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
With this PR: huggingface/optimum-executorch#207 we are adding a new method "sampler" to ASR models, alongside with "encoder" and "text_decoder". The flow becomes: if temperature is 0 and sampler method is available, run that method. Otherwise still go with the old path. This change should largely improve the performance on CUDA since we don't have to copy logits from device to CPU for sampling purpose.
Benchmark result on RTX 5080: