5

I'm trying to deploy a model version in google cloud ai-platform. I successfully create the model (called MyModel), but when I try to deploy a version of the model from the command line, it prompts the next error:

Create Version failed. Bad model detected with error: "Failed to load model: Unexpected error when loading the model: problem in predictor - ImportError: cannot import name 'collections_abc' from 'six.moves' (unknown location) (Error code: 0)"

I actually created the model without troubles 2 weeks ago. I also was having the same problem with other models and I solved it with the next libraries in the setup:

  • google-auth==1.18.0
  • google-cloud-storage==1.31.2
  • google-api-core==1.22.2

The way the model is created is to look at the model dist from a storage bucket (I uploaded it from the command line with no problems):

Steps to create the model

  1. Create targz file that will be uploaded to the cloud storage bucket next to the model
    • python setup.py sdist --formats=gztar
  2. Upload tarball to gcloud storage
    • gsutil cp dist/my_custom_code-0.1.tar.gz gs://my-bucket
  3. Create ai-platform model
    • gcloud beta ai-platform models create MyModel --regions my-region --enable-logging --enable-console-logging
  4. Create version model (v1)
    • gcloud beta ai-platform versions create v1 --model MyModel --runtime-version 1.15 --python-version 3.7 --origin gs://my-bucket/model/ --package-uris gs://my-bucket/my_custom_code-0.1.tar.gz --prediction-class predictor.MyPredictor

What have I tried

  • Create the version with the last google libraries versions
  • Create the version with the next libraries: google-auth==1.18.0, google-cloud-storage==1.31.2, google-api-core==1.22.2

I suppose it is something related with google libraries and their updates in the versions. I am open to all suggestions. Thanks in advance.

Francisco Javier Gomez Sánchez
  • 53
  • 4

0 Answers0