Skip to content

[Accuracy diff No. 110, 111] Fix accuracy diff for paddle.nn.functional.binary_cross_entropy, paddle.nn.functional.binary_cross_entropy_with_logits API - #267

Closed
NKNaN wants to merge 4 commits into
PFCCLab:mainfrom
NKNaN:bce-fix
Closed

[Accuracy diff No. 110, 111] Fix accuracy diff for paddle.nn.functional.binary_cross_entropy, paddle.nn.functional.binary_cross_entropy_with_logits API#267
NKNaN wants to merge 4 commits into
PFCCLab:mainfrom
NKNaN:bce-fix

Conversation

@NKNaN

@NKNaN NKNaN commented Jun 15, 2025

Copy link
Copy Markdown
Contributor

目前 paddle.nn.functional.binary_cross_entropy, paddle.nn.functional.binary_cross_entropy_with_logits 对应的反向算子只计算 x_grad 不计算 label_grad，而 torch 计算 x_grad 和 target_grad (label_grad)：
image
image

第一次回测结果：
image

paddle.nn.functional.binary_cross_entropy_with_logits 能全部通过
paddle.nn.functional.binary_cross_entropy 有两个 case 未通过：

paddle.nn.functional.binary_cross_entropy(Tensor([195840, 80],"float16"), Tensor([195840, 80],"float16"), reduction="none", )
paddle.nn.functional.binary_cross_entropy(Tensor([87040, 80],"float16"), Tensor([87040, 80],"float16"), reduction="none", )

对这两个 case 二次回测结果：
image

@luotao1 luotao1 mentioned this pull request Jun 16, 2025
Closed
cangtianhuang
cangtianhuang requested changes Jun 16, 2025
View reviewed changes

@cangtianhuang cangtianhuang left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

如果修改可以全部 pass 的话，请将所有会报 [not compare] 错误的配置，从：tester/api_config/5_accuracy/accuracy_7.txt 移动至 tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt，便于后续复测以及修复～

@NKNaN

NKNaN commented Jun 16, 2025

Copy link
Copy Markdown
Contributor Author

如果修改可以全部 pass 的话，请将所有会报 [not compare] 错误的配置，从：tester/api_config/5_accuracy/accuracy_7.txt 移动至 tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt，便于后续复测以及修复～

好的

@cangtianhuang

cangtianhuang commented Jun 16, 2025

Copy link
Copy Markdown
Collaborator

@NKNaN 以及 tester/api_config/5_accuracy/accuracy_gpu_error.txt 中的所有 [not compare] 出错配置，也复制一份到 tester/api_config/5_accuracy/accuracy_gpu_error_grads_diff.txt 中，保留原有的便于复测 ，辛苦同学

@NKNaN

NKNaN commented Jun 16, 2025

Copy link
Copy Markdown
Contributor Author

好的，tester/api_config/5_accuracy/accuracy_gpu_error.txt 中的所有 [not compare] 出错配置 是 15857 个，在上面回测通过；tester/api_config/5_accuracy/accuracy_7.txt 是 602 个，回测全部通过：
image

@cangtianhuang

cangtianhuang commented Jun 16, 2025

Copy link
Copy Markdown
Collaborator

不好意思同学，是我描述有点问题，最终应该这样处理：

  • accuracy_7.txt 在修复前能正常 pass，我们就不管它
  • accuracy_gpu_error.txt 修复好的配置 复制 一份到 accuracy_gpu_error_grads_diff.txt 中，保留原有的😊

@NKNaN

NKNaN commented Jun 16, 2025 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

感谢！
accuracy_7.txt 的 602 个在修复前可以正常pass
image

@NKNaN

NKNaN commented Jun 16, 2025 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

我 reset 一下

@cangtianhuang

cangtianhuang commented Jun 16, 2025

Copy link
Copy Markdown
Collaborator

我 revert 一下

好滴，看上去误删了很多，这个应该是 git 索引的问题，文件偏大就会有这个问题😂

@NKNaN NKNaN mentioned this pull request Jun 16, 2025
Merged
@NKNaN

NKNaN commented Jun 16, 2025

Copy link
Copy Markdown
Contributor Author

我 revert 一下

好滴，看上去误删了很多，这个应该是 git 索引的问题，文件偏大就会有这个问题😂

重新提了一个 pr

@cangtianhuang

cangtianhuang commented Jun 16, 2025

Copy link
Copy Markdown
Collaborator

this pr can be closed safely

@NKNaN NKNaN closed this Jun 16, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@cangtianhuang cangtianhuang cangtianhuang requested changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@NKNaN @cangtianhuang