[inductor] Enable coordinate descent tuning with max-autotune #136867
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/136867
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit a9b04d0 with merge base 22e19bd ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot rebase |
|
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
|
Successfully rebased |
There was a problem hiding this comment.
Would this be a good opportunity to add opt_level as a kwarg ? We just had reports internally for a model that doubled compilation time with coordinate descent tuning.
I think we could also roll this out by default in max-autotune by only targeting the relatively more expensive kernels.
|
In my testing the added compile time was small compared to matmuls. I think we should embrace the "max" in "max-autotune". |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
4 participants
Stack from ghstack (oldest at bottom):
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang