Oracle Cloud Infrastructure Documentation

Custom Models

Troubleshoot common problems with custom models in Document Understanding.

Job Run Couldn't be Started because of Service Issues

A job run doesn't start and the following error message is shown: Job run could not be started due to service issues. Please try again later.

This is a service issue.

  1. Wait a few minutes and try again.
  2. If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Parsing Records Returns an Error

While parsing records, the following error is returned: Please ensure you are not using an empty dataset, and that your dataset matches the type of model you are trying to train.

The dataset records might be corrupted in one of the following ways:
  • Missing a label
  • The labeled data has a different annotation class to the custom model you're trying to train.
  • The document is smaller than 32 x 32 pixels.
  • A file is password-protected.
  • A file is corrupt.
.
  1. Check that the documents are labeled and label any that aren't.
  2. Confirm that the labeled data has the same annotation class as the custom model you're trying to train.

    Key value extraction models consume labeled data of a key-value annotation class, and document classification models consume labeled data of single-class classification annotation class.

  3. Remove any documents with dimensions smaller than 32 x 32 pixels.
  4. If any files are password-protected, remove the protection.
  5. Remove any corrupted files.

Not Enough Samples to Split Dataset into Train, Validation, and Test Sets

Training a dataset fails with the error: Not enough samples to split dataset into Train, Validation and Test sets. Ensure the required number of samples for each class are in the dataset.

The error might also include information the number of records for each label. For example: {'a': {'val': 1, 'test': 1, 'train': 3}, 'b': {'val': 0, 'test': 0, 'train': 3}}

There aren't enough valid records for each label. In the example error, there aren't enough records for label "b" (only three documents are labeled). Label two more records with label b to resolve the issue.

Add more records for the labels that don't have enough.

Error while Loading the Visual Model

While loading the model you get the following error: Error while loading the visual model. Please check the base model version.

This is a service issue.

If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Compute Resource not Available to Train Model

While training a custom model, you get the following error: Compute resource not available to train model. Please try again later.

This is a service issue.

  1. Wait a few minutes and try again.
  2. If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Create Model Failed

Creating a custom model fails with the following error: Create Model failed because the given training dataset with OCID does not exist, or has too few Records or Annotations.

The dataset is incomplete or doesn't exist.

  1. Make the dataset available.
  2. Give it the required number of records and annotations.

Invalid DLS Dataset Annotation Type

When using a dataset from Data Labeling, you get the following error: Invalid dls dataset annotation type. There is a mismatch in the model type and dls dataset annotation type. Please select a dataset with <> annotation type for <> model type.

Status Code: 400

The dataset format and model type don't match.

Correct the dataset format and model type so they match.

Error Occurred during Dataset File () Validation

When using a custom model, you get the error:

Error occurred during dataset file () validation. Invalid annotationFormat: <> Please check annotationFormat inside dataset file. Expected annotationFormat is <>

Status Code: 400

The dataset format and model type don't match.

Correct one of the dataset format or the model type, so that they match.

Fail to Export the Annotation File for DLS Dataset

When using a dataset from Data Labeling, you get the following error: Fail to export the annotation file for dls dataset.

This is a service issue.

  1. Wait a few minutes and try again.
  2. If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Failed in DLS Snapshot Wait Step

When using a custom model based on a Data Labeling dataset, you get the following error: Failed in DLS snapshot wait step.

This is a service issue.

  1. Wait a few minutes and try again.
  2. If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Create Model Failed while Querying Job Shapes

While creating a custom model, you get the error: Create model failed while querying job shapes.

This is a service issue.

  1. Wait a few minutes and try again.
  2. If the job run still fails, send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

Model-Count or Project-Count Limit Exceeded

While using a custom model, you get the error: model-count/project-count limit exceeded.

The limits for models or projects have been exceeded.

  1. Rasie a request to increate the model or project count for Document Understanding.
  2. Delete any models or projects that are no longer needed.

Deployment Failed for Model

Using a custom model, you get the error: Deployment failed for model <model_ocid>.

This is a service issue.

Send an email to contact_document_understanding_ww_grp@oracle.com with the model ocid and region details for further help.

All Classes should have at Least {MIN_REQUIRED_SAMPLES} Samples for Data Split

When using a custom model, you get the error: All classes should have at least {MIN_REQUIRED_SAMPLES} samples for data split. Classes ["a", "b"] have less than required number of samples.

The input dataset is missing records for the labels mentioned. In the example, the labels a and b are missing records.

Add the necessary records to the labels mentioned in the error message.