1

I am trying to test a model that I onboarded to the Acumos platform, and upon downloading the .tar file and loading the image in docker, running the image in docker fails. The reported cause is...

raise IOError("File %s does not exist." % filename) OSError: File model/tmp79b3brlw\graph.meta does not exist.

This is when I onboard from a windows machine. However, when I try running a .tar file that was already on the marketplace (sklearn iris example), the docker runs fine. I went to the acumos soups to nuts tutorial site, and copy and pasted the example code for the iris example and ran it, and was able to onboard it fine. However, when I run that onboarded version as a model user, it fails with the above exception (graph.meta missing). I noticed a backslash and forward slash in the path to the graph.meta, and wonder if windows encoding in UTF-8, and linux encoding in unicode may lie at the heart of the issue? Is it possible to onboard a tensorflow model from windows? Any help is greatly appreciated!

Andrew James
  • 372
  • 1
  • 8
  • The forward/backward slashes definitely look suspicious. Your model is Python? Can you share the code somewhere? Also this answer about on-boarding a TF model may help: https://stackoverflow.com/questions/51681055/uploading-tf-and-possibly-txt-file-dependencies-to-acumos – chrisinmtown Aug 05 '18 at 00:51
  • Problem is tracked in https://jira.acumos.org/browse/ACUMOS-1515 – chrisinmtown Aug 10 '18 at 20:30

0 Answers0