Hi! I think the issue is that when you create the Dockerfile, you are using /workflow/kubeflow-mnist. However, when you use the VolumeOp and mount it on /workspace, the previous contents in /workflow/kubeflow-mnist is overridden. That's why it cannot find the files. In my example, I git clone-ed the repository into the VolumeOp. Hope this helps! Ben
…
On Thu, Sep 24, 2020 at 4:28 PM bruceche11 @.***> wrote: hello, i learn you blog https://www.jianshu.com/p/b4f40d4c5f4e and try the fashion-mnist example, and here comes the issue: [image: image] https://user-images.githubusercontent.com/16239670/94120164-6744ff00-fe82-11ea-82e1-17cbb719f89e.png here is my Dockerfile [image: image] https://user-images.githubusercontent.com/16239670/94120339-922f5300-fe82-11ea-82e7-0513a9ebde34.png my pipeline.py [image: image] https://user-images.githubusercontent.com/16239670/94120456-b428d580-fe82-11ea-962f-f10a7fa67084.png [image: image] https://user-images.githubusercontent.com/16239670/94120520-cc98f000-fe82-11ea-8a7f-7a0b8615cd03.png the preprocessing.py and the train.py is the same as yours. Thanks in advance — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAGSINGR7MEIXY67WVNGCXDSHL7KRANCNFSM4RYACRUA .

Sorry to reply late.
Yes, you are right, that is the reason. I have figured it out. Thank you.