New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Freeze fuse two mms #111232
Freeze fuse two mms #111232
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/111232
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 0d8a9d4 with merge base 543a763 (): FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
UNSTABLE - The following job failed but was likely due to flakiness present on trunk and has been marked as unstable:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Improves llama_v2 perf locally from 1.48x -> 1.55x. A good future rewrite would be to unify the freezing batching with the other batching rules that yanboliang & co were working on. I want to wait for the forthcoming pre-dispatch changes to settle down first. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler [ghstack-poisoned]
Improves llama_v2 perf locally from 1.48x -> 1.55x. A good future rewrite would be to unify the freezing batching with the other batching rules that yanboliang & co were working on. I want to wait for the forthcoming pre-dispatch changes to settle down first. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler [ghstack-poisoned]
batching rules you say?? |
To be more clear, this batches horizontal mms that share a single input where the other inputs are parameters, and constant folds the concat of the params in freezing. |
self.t1 = torch.nn.Parameter(torch.rand(10, 10)) | ||
self.t2 = torch.nn.Parameter(torch.rand(10, 10)) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you add a test where the sizes of t1/t2 aren't the same?
Improves llama_v2 perf locally from 1.48x -> 1.55x. A good future rewrite would be to unify the freezing batching with the other batching rules that yanboliang & co were working on. I want to wait for the forthcoming pre-dispatch changes to settle down first. cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 aakhundov ColinPeppler [ghstack-poisoned]
@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 |
ghstack-source-id: e2f6c418861608b399ca08733dec4e1c38e79ba1 Pull Request resolved: pytorch/pytorch#111232
Improves llama_v2 perf locally from 1.48x -> 1.55x. A good future rewrite would be to unify the freezing batching with the other batching rules that @yanboliang & co were working on. I want to wait for the forthcoming pre-dispatch changes to settle down first. Pull Request resolved: pytorch#111232 Approved by: https://github.com/Chillee
Stack from ghstack (oldest at bottom):
Improves llama_v2 perf locally from 1.48x -> 1.55x.
A good future rewrite would be to unify the freezing batching with the other batching rules that @yanboliang & co were working on. I want to wait for the forthcoming pre-dispatch changes to settle down first.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @aakhundov @ColinPeppler