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
nn.Linear: dispatch to bsr_dense_mm for half and bfloat16 #94825
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/94825
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit bc7ed5c: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
ghstack-source-id: 6790e06cf0ddf3b33831ebba8cc638f8956f631a Pull Request resolved: #94825
[ghstack-poisoned]
ghstack-source-id: 133a63c1ceea38d9d24e8e3d5d8622ed139284a1 Pull Request resolved: #94825
[ghstack-poisoned]
ghstack-source-id: bf2abe355707ed10c0fd913a002a3a8cbb7f2e9f Pull Request resolved: #94825
[ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
@cpuhrsch , let's try this one as well? |
@nikitaved - Were you able to locally reproduce the deadlock with CUDA 11.6? |
No, I was not. Looks like the issue might have been fixed on the Triton's side. The first PR in the stack does run the tests though with no issues so far, so we could give it some more grace time? |
@nikitaved - you can install a different (older) version of Triton via pypi and PyTorch will pick up on it. For example
If you do this with a build based on 11.6, does it reproduce the issue? Is there any way we can reconstruct the environment under which you've observed these failures? I'm worried to land something this critical that we know has had issues in the past without resolving them. |
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
ghstack-source-id: 90a98a5b496a6da02080d1fae976f1c3d7de26bd Pull Request resolved: #94825
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
ghstack-source-id: f8650b35ad68727fccafa65a8c3ba87f1f98ad2a Pull Request resolved: #94825
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
ghstack-source-id: 68f44145c6df2794fb557826389a0d94db132182 Pull Request resolved: #94825
@cpuhrsch , I was not able to fail it, so, I guess, this one should be good to go. |
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
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.
This is amazing. Triton in eager mode!
Unfortunately there's still a lint error left, but otherwise let's give this another go. |
cc alexsamardzic pearu cpuhrsch amjames bhosmer [ghstack-poisoned]
@pytorchbot merge -g |
❌ 🤖 pytorchbot command failed:
Try |
@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 |
Pull Request resolved: #94825 Approved by: https://github.com/albanD, https://github.com/cpuhrsch
…2] (pytorch#102660) Test was originally skipped in pytorch#98462 Not sure why it was removed in pytorch#94825 Now the test hits CUDA illegal memory access on H100 again after pytorch#101163 Pull Request resolved: pytorch#102660 Approved by: https://github.com/zou3519
Stack from ghstack (oldest at bottom):
cc @alexsamardzic @pearu @cpuhrsch @amjames @bhosmer