fix: correct DDP handling and backward gradients - #214
Merged
Conversation
chen2021673
commented
Aug 31, 2026
chen2021673
commented
Aug 31, 2026
chen2021673
force-pushed
the
fix/backend-independent-correctness
branch
from
September 2, 2026 09:11
0c2b716 to
26d2402
Compare
Contributor
Author
|
精度测试： 这里精度出现偏差主要是因为 layernorm 和 linear kernel 的修复影响了 GPT 测例的精度。以 gpt2_1 测例为例，修复后精度与pytorch更加接近： 性能测试： |
kilinchange
self-requested a review
September 11, 2026 02:35
kilinchange
requested changes
Sep 11, 2026
|
|
||
| virtual int GetGroupRank(int global_rank) const; | ||
|
|
||
| Device::DeviceType backend() const { return backend_; } |
chen2021673
force-pushed
the
fix/backend-independent-correctness
branch
from
September 11, 2026 05:24
519872c to
22bf5b5
Compare
- honor DDP reduction configuration and validate process group devices - fix elementwise and CUDA linear backward correctness - remove the unsupported distributed optimizer test - clean up related build diagnostics and test documentation
Contributor
Author
复测精度测试： gpt2 测例精度得到修正，llama3 测例不变。 性能测试： 视为波动。 |
kilinchange
approved these changes
Sep 11, 2026
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Fix several backend-independent correctness issues in DDP, autograd, and CUDA linear backward.
Changes
average_in_collectiveacross DDP reduction pathsExpbackward and avoid aliased gradients inAddbackward