-
Notifications
You must be signed in to change notification settings - Fork 25.6k
[ROCm] Autocast RNN Support #121539
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
[ROCm] Autocast RNN Support #121539
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/121539
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit 6863e63 with merge base 75bb049 ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Please seek CI approval before scheduling CIFlow labels |
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.
Sounds good!
Will let @jeffdaily do a final review and trigger the merge
@pytorchbot merge |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
Didn't find following labels among repository labels: release notes: Autocast support for MIOpen RNNs |
@pytorchbot merge |
Didn't find following labels among repository labels: release notes: |
@pytorchbot merge -f "Bot is confused about labels. CI is green" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Test was enabled in the upstream pytorch#121539 But failed with "AssertionError: Tensor-likes are not close! Mismatched elements: 30 / 30 (100.0%)" in release/2.3 and rocm6.3_internal_testing on MI300 and passed on MI210 Pass with AMD_LOG_LEVEL=3 on MI300 Need to dig deeper Skip TestCuda::test_autocast_rnn as in other releases (https://ontrack-internal.amd.com/browse/SWDEV-468117) release/2.1 https://github.com/ROCm/pytorch/blob/073f5dcaebc61b69a71de8918e574053bf95c40b/test/test_cuda.py#L2023 release/2.2 https://github.com/ROCm/pytorch/blob/a3208baad2f2d537f10362fe9ca30f60ff5aea21/test/test_cuda.py#L2047 Co-authored-by: root <root@banff-cyxtera-s80-2.amd.com>
Fixes #116361
Implements Autocast wrapper for miopen rnn's
cc @jeffdaily @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang