("Failed to load the model due to bad model data. tags: ['serve']\nNo op named HashTableV2 in defined operations.", 1)

Question

I'm using GCP with GCS and GCMLE APIs to train my model and to predict on the cloud. I'm working on the census dataset. And the goal is to predict a profile's income. this is the github link of codes and scripts https://github.com/SofiaAmel/censusTest Actually I am discovering this tool, I'm struggling with prediction, I'm testing with one profile as .json input , the job preiction runs succussfully, but it occurs this error ***

("Failed to load the model due to bad model data. tags: ['serve']\nNo op named HashTableV2 in defined operations.", 1)

---

ps : I'm using TF version 1.4 , if you can help don't hesitate ! Thanks in advance

Answer 1

This error message indicates that you have chosen a lower version of TensorFlow for serving the model than you used to train the model and the HashTableV2 op was introduced in a later version.

To fix this, be sure to specify the correct runtime version when deploying the model, e.g.,

gcloud ml-engine version create ${VERSION} --model ${MODEL} --runtime-version=1.4

At this time, it is not possible to specify the runtime version when using Cloud Console. We hope to add the feature shortly.