-
Notifications
You must be signed in to change notification settings - Fork 4.2k
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
Upgrade ncnn cause output dimension wrong #4875
Comments
please provide a minimal model, pytorch module code or torchscript, for reproducing your issue |
Here is the model: 链接:https://pan.baidu.com/s/14vmHbVxTtuXeuovSEp933g Also I've found on the same versions using C++, the result is different but the shape is the same. |
After analyzing the model more, I've found that the input shape into the gemm layer is the same between NCNN versions. However, the output shape if the gemm layer is different, with previous versions having the output (256,1,1,1) and the newer version with (256,4,1,1). Here is the C++ 14 code if anyone wants to try:
|
Thank you for providing the code and model for testing. The bug has been identified and I am currently submitting the pull request |
Thank you, it's working now as of PR #4890 |
After upgrading NCNN from 20220729 to 20230517, the output from my model is incorrect
Here's the code for running the model:
With the older version the output shape is
1,1,256
, the newer version the shape is1,8,256
The text was updated successfully, but these errors were encountered: