Skip to content
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

转换ConvNext预训练模型时,提前退出,无报错信息 #1200

Closed
ssyj911 opened this issue Apr 30, 2024 · 4 comments · Fixed by #1204
Closed

转换ConvNext预训练模型时,提前退出,无报错信息 #1200

ssyj911 opened this issue Apr 30, 2024 · 4 comments · Fixed by #1204

Comments

@ssyj911
Copy link

ssyj911 commented Apr 30, 2024

Describe the bug
将ConvNext预训练模型转kmodel时,nncase 编译器输出:"2.Optimize target independent..." 后退出。无任何其他提示

To Reproduce
预训练模型可以在以下地址获取:
https://dl.fbaipublicfiles.com/convnext/convnext_tiny_1k_224_ema.pth

Origin model and code
按照 nncase 提供的 编译float32 onnx模型 代码进行 kmodel 进行转换

Environment (please complete the following information):

  • OS: [ Ubuntu20.04]
  • nncase version 1.9.0-3473131
  • DL Framework [python 3.8]

Additional context
尝试将nncase安装包中的 nncase-1.9.0/examples/yolox/model/yolox_nano_224.onnx 进行转换,未发现问题

@curioyang
Copy link
Member

@ssyj911 请直接提供onnx模型

@ssyj911
Copy link
Author

ssyj911 commented Apr 30, 2024

@ssyj911 请直接提供onnx模型

这个太大了,上传到百度云盘啦:链接: https://pan.baidu.com/s/1MEIbKCnJama3k1AjjSn4uw?pwd=ymgf 提取码: ymgf

@curioyang
Copy link
Member

@ssyj911 编译的问题已经修复了,但是这个模型里有大量的CPU算子layernorm,在实际推理的时候速度很慢,且510 在设计上存在一定的精度损失。
如果在510上部署实际项目,建议更换模型
或者可以尝试在230上部署,int16量化,230对部分CPU算子是有V扩展优化的,且不太容易出现精度损失

@curioyang curioyang linked a pull request May 14, 2024 that will close this issue
@ssyj911
Copy link
Author

ssyj911 commented May 14, 2024 via email

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants