Skip to content

phycrnet 在A100平台精度结果与官方差距大 #1144

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.

Sign up for GitHub

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

Jump to bottom
Open
yuchunliu-metax opened this issue Apr 26, 2025 · 3 comments
Open

phycrnet 在A100平台精度结果与官方差距大 #1144

yuchunliu-metax opened this issue Apr 26, 2025 · 3 comments

Comments

@yuchunliu-metax
Copy link
Contributor

bug 描述 bug description

软件版本：paddle：2.6.2 , PaddleScience：commit ab417c8
测试案例：PaddleScience/examples/phycrnet
测试环境：NVIDIA A100-PCIE-40GB

A100 训练日志 ： train_log.txt
A100 推理结果 ：a-RMSE: 2.0e-2

Image

官方推理结果：a-RMSE: 3.20e-3

Image

A100 推理结果与官方推理结果差距较大！

其他补充信息 Additional Supplementary Information

No response
@yuchunliu-metax yuchunliu-metax changed the title phycrnet 在A100平台训练精度结果与官方差距大 phycrnet 在A100平台精度结果与官方差距大 Apr 26, 2025
@wangguan1995
Copy link
Contributor

wangguan1995 commented Apr 27, 2025 •
edited
Loading

看上图的结果(应该是paddle?)：

  • 误差区间约为[5.0e-3, 2.0e-2]
  • 时间区间为[0, 4]

看下图的结果(应该是torch？)：

  • 误差区间约为[3.20e-3, 1.0e-2]
  • 时间区间为[0, 4]

Image

@wangguan1995
Copy link
Contributor

wangguan1995 commented Apr 27, 2025 •
edited
Loading

请问能不能提供下：torch 和 paddle的checkpoint方便复现呢？？

@yuchunliu-metax
Copy link
Contributor Author

yuchunliu-metax commented Apr 28, 2025 •
edited
Loading

A100环境测试：
第一张图：paddle 推理 A100训练结果， 使用权重 latest.pdparams
第二张图：paddle 推理 官方的权重， 使用权重 phycrnet_burgers.pdparams
上传了一个压缩包，包含latest.pdparam、phycrnet_burgers.pdparams

流场高分辨率重构 PhyCRNet.zip

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@wangguan1995 @yuchunliu-metax