Help Center> ModelArts> FAQs> Model Management> Compressing and Converting Models> What Can I Do If Converting a TensorFlow Model Failed?
View PDF

What Can I Do If Converting a TensorFlow Model Failed?

Symptom

Converting a model created using TensorFlow failed, as shown in the following figure.

Figure 1 Model conversion failed

Solution

Perform the following operations to locate the fault:

  1. Check whether the current account has been granted with permissions on Conversion Input Path and Conversion Output Path in the model conversion task.
    Figure 2 OBS paths in a model conversion task

    Check method:

    1. Log in to the OBS management console, locate the target OBS bucket in the bucket list, and click the bucket name to go to the bucket details page.
    2. In the left navigation pane, choose Permissions > Bucket ACLs and check whether the current account has the read and write permissions.
      • If yes, go to the next step.
      • If no, use another OBS bucket or contact the owner of the OBS bucket to grant the permissions.
      Figure 3 Viewing permissions
  2. Check whether the model to be converted contains any unsupported operators.

    Model conversion applies only to the Caffe and TensorFlow operators with operator requirement met. Check your model according to the operator list in the Ascend community and ensure that the operators used in your model are supported.

    When the preceding causes are excluded, run the model conversion task again.