Update test_gpt_pp baselines for cross_entropy precision alignment#702
Open
zrr1999 wants to merge 1 commit intoPaddlePaddle:release/0.2from
Open
Update test_gpt_pp baselines for cross_entropy precision alignment#702zrr1999 wants to merge 1 commit intoPaddlePaddle:release/0.2from
zrr1999 wants to merge 1 commit intoPaddlePaddle:release/0.2from
Conversation
…gnment Update MD5 baseline checksums in test_gpt_pp.py to match the new gradient values produced by Paddle PR #78044, which aligns cross_entropy precision with PyTorch under FLAGS_use_accuracy_compatible_kernel. The cross_entropy grad kernel now uses AccT (MPType) arithmetic instead of native T arithmetic, which changes intermediate gradient values and propagates through the pipeline parallel backward pass. Loss MD5 remains unchanged (bce3fed95247f1b7a165e32b33d6fca7). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1 task
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
test_gpt_pp.pyto match new values from Paddle PR [PAA] Align cross_entropy precision for hard-label 2D reduction under FLAGS_use_accuracy_compatible_kernel Paddle#78044AccT(MPType) arithmetic, which changes gradient values propagated through pipeline parallel backward passbce3fed95247f1b7a165e32b33d6fca7)devPR: #703
Context
Paddle PR #78044 aligns
cross_entropyprecision with PyTorch underFLAGS_use_accuracy_compatible_kernel. The grad kernel change (usingAccTinstead of nativeTarithmetic) affects gradient computation even in the default path, causing thetest_gpt_ppmulti-card test to fail with mismatched baseline checksums.New baseline values are taken from the CI run output of PR #78044.
Test plan
🤖 Generated with Claude Code