AI Builder: Troubleshooting Missing Training Documents in Your Document Processing Model

Table of Contents

This article addresses a common issue encountered by users of AI Builder’s document processing models: the absence of training document previews on the model details page. Understanding why these documents might be missing and how to rectify the situation is crucial for effectively managing and improving your AI models. This guide will walk you through the potential causes and provide clear resolutions to ensure you can access and utilize your training data as intended.

Understanding the Symptom: Training Document Preview Not Displayed¶

When you successfully train a document processing model within AI Builder, a key feature is the display of a preview of one of your training documents on the model details page. This preview serves as a visual confirmation and allows you to quickly review the document that was used for training. However, in some instances, this expected preview fails to appear. Instead of seeing a thumbnail or a representation of your training document, you might find a blank space, a generic icon, or simply no indication that a training document is available for preview.

This missing preview can be perplexing and raise concerns about the integrity of the training process. You might wonder if the model was trained correctly, if the data was properly ingested, or if there is a problem with the platform itself. While the absence of a preview can indicate an underlying issue, it’s often related to specific, resolvable causes. It’s important to systematically investigate these potential causes to restore the expected functionality.

The symptom is specifically related to the preview on the model details page. This means the issue is not necessarily about the model failing to train or process documents in general. It’s about the visual representation of the training data within the AI Builder interface. The model might still be functioning as intended, but the lack of preview hinders your ability to quickly validate and understand the training data it used.

Potential Causes for Missing Training Documents¶

Several factors can lead to the training document preview not being displayed. These causes generally fall into categories of permissions, model ownership, and model origin (specifically, whether the model was imported from another environment). Let’s explore each of these in detail:

1. Insufficient Permissions to View Training Documents¶

Access to resources within Power Platform, including AI Builder models and their associated data, is governed by permissions. It’s possible that your current user account lacks the necessary permissions to view the training documents associated with the document processing model. These permissions are typically managed at the environment level or specifically assigned to individual resources.

Permissions in Power Platform are role-based. Different roles grant varying levels of access and control. For instance, a user with a basic user role might have limited access compared to a user with an environment administrator role. Viewing training documents is likely to require a certain level of access that allows you to interact with the model’s training data. If your assigned role doesn’t include this level of access, the preview will be suppressed, even if the model itself is functioning correctly.

To determine if this is the cause, you need to investigate your current user role within the Power Platform environment where the AI Builder model resides. You also need to understand what permissions are required to view training documents within AI Builder. This often involves consulting your organization’s Power Platform administrator or reviewing the documentation related to AI Builder security roles.

2. Not Being the Model Owner¶

Model ownership is another crucial concept within AI Builder. The owner of a model typically has full control and access to all aspects of the model, including its training data. If you are not the owner of the document processing model in question, your access might be restricted, potentially preventing you from viewing the training document preview.

Even if you have general permissions within the environment, model ownership adds an additional layer of access control. The creator of the model is usually designated as the initial owner. Ownership can be transferred or shared, but if you are not explicitly granted owner or co-owner status, your ability to interact with certain model features, such as viewing training data previews, might be limited.

To verify if ownership is the issue, you need to identify the current owner of the model. This information is usually available within the model details page in AI Builder, although it might require specific permissions to view. If you discover that you are not the owner, you will need to coordinate with the model owner to either gain ownership or co-owner permissions, or to request them to investigate the missing preview issue.

3. Model Imported from Another Environment¶

A common practice in Power Platform environments is to export and import solutions, which can include AI Builder models. However, when a document processing model is imported from one environment to another, a critical aspect to understand is that the original training data is not migrated along with the model itself. This is a design decision, likely due to data privacy, storage efficiency, and environment isolation considerations.

When a model is imported, only the model definition, configuration, and potentially some metadata are transferred. The actual documents used to train the model in the source environment remain in that source environment. Consequently, when you view the imported model in the destination environment, there is no associated training data readily available to generate the preview.

This situation is a common cause for missing training document previews, especially in development, testing, and production environment scenarios where models are frequently moved across environments. If you know that your model was imported, this is highly likely to be the reason for the missing preview. In this case, the missing preview is not an error, but rather an expected outcome of the model import process.

It is important to note that while the original training data is not imported, you can retrain the imported model in the new environment using new training documents. However, this will create a new training dataset within the destination environment, and the preview will then be based on these new training documents, not the original ones from the source environment.

Resolving the Missing Training Document Preview Issue¶

The resolution for the missing training document preview depends directly on the identified cause. Here are the recommended steps to take for each potential scenario:

Resolution for Permission Issues¶

If you suspect that insufficient permissions are the cause, the first step is to verify your current permissions. This might involve:

1. Checking your assigned roles: Within the Power Platform admin center or environment settings, you can typically view the roles assigned to your user account. Look for roles related to AI Builder or general Power Platform access.
2. Consulting documentation: Review the official AI Builder documentation to understand the specific permissions required to view training documents. This documentation will outline the necessary roles and privileges.
3. Contacting your administrator: If you are unsure about your permissions or how to check them, reach out to your Power Platform administrator or IT support team. They can review your permissions and grant you the necessary access if required.

Once you have confirmed that your permissions are indeed insufficient, the resolution is to request the necessary permissions. This typically involves:

1. Identifying the required role: Determine the specific role or set of permissions that grant access to view training documents in AI Builder.
2. Submitting a permission request: Follow your organization’s process for requesting access to systems and applications. This might involve submitting a formal request form, contacting your administrator directly, or using an internal IT support portal.
3. Providing justification: Clearly explain why you need access to view training documents. Highlighting your role in model management, improvement, or validation can strengthen your request.

After your permissions are updated, refresh your AI Builder environment and navigate back to the model details page. The training document preview should now be visible, assuming permissions were indeed the root cause.

Resolution for Model Ownership Issues¶

If you are not the owner of the model, and this is preventing you from seeing the training document preview, there are two main resolution paths:

1. Request Co-owner Permissions: The simplest solution is often to request co-owner permissions from the current model owner. Co-ownership grants you many of the same rights as the owner, including access to training data and previews. To request co-ownership:

   • Identify the model owner: Find out who the current owner is, usually visible within the model details.
   • Contact the owner: Reach out to the owner and explain your need for co-owner access.
   • Request co-ownership within AI Builder: The model owner can grant co-owner permissions directly within the AI Builder interface.

2. Request Ownership Transfer (If Appropriate): In some cases, it might be more appropriate to transfer full ownership of the model to you, especially if the original owner is no longer actively involved in managing the model or if you are taking over responsibility for it. Ownership transfer is a more significant change than granting co-owner permissions. To request ownership transfer:

   • Discuss with the current owner: Have a conversation with the current owner to discuss the possibility of transferring ownership.
   • Understand organizational policies: Check if your organization has specific policies or procedures regarding model ownership and transfer.
   • Initiate ownership transfer within AI Builder: The current model owner can initiate the ownership transfer process through the AI Builder interface.

Once you have either gained co-owner permissions or ownership, revisit the model details page. The training document preview should now be accessible, assuming ownership was the limiting factor.

Resolution for Imported Models¶

If your document processing model was imported from another environment, and you are missing the training document preview, it’s important to understand that this is the expected behavior. The original training data is simply not present in the destination environment. In this scenario, there is no “resolution” in the sense of recovering the original preview.

However, you can still take steps to work with the imported model effectively:

1. Retrain the Model (Recommended): The best approach is to retrain the imported model in the new environment using a new set of training documents. This will create a new training dataset within the current environment, and a preview will be generated based on these new documents after successful retraining. Retraining also ensures the model is adapted to any potential differences between the source and destination environments.

2. Access Original Training Data in Source Environment (If Possible): If you need to review the original training documents, you will need to access the source environment from which the model was exported. Navigate to the model in the source environment, and the training document preview should be visible there (assuming permissions and ownership are correctly configured in the source environment). However, directly using the original training data in the destination environment is generally not possible due to the import process limitations.

In summary, for imported models, the focus should be on retraining the model in the new environment to establish a new training dataset and enable the document preview feature. Attempting to recover the original preview from an imported model is generally not feasible.

Importance of Training Document Preview¶

While the missing training document preview might seem like a minor visual issue, it’s connected to important aspects of model management and quality control. Being able to preview the training documents offers several benefits:

• Validation of Training Data: The preview allows you to quickly visually confirm that the correct documents were used for training. You can verify that the documents are of the expected type, format, and content. This helps catch errors early in the model development process.
• Quality Check of Training Set: By reviewing training documents, you can assess the overall quality and representativeness of your training dataset. Are there any inconsistencies? Are the documents sufficiently diverse? Does the dataset cover the range of document variations the model will encounter in production?
• Understanding Model Behavior: Examining the training documents provides context for understanding how the model is likely to perform. You can anticipate potential strengths and weaknesses based on the characteristics of the data it was trained on.
• Troubleshooting Model Performance: If the model performs poorly in production, reviewing the training documents can provide clues about potential issues. Perhaps the training data was insufficient, biased, or didn’t accurately represent real-world scenarios.
• Collaboration and Communication: The preview can be a useful tool for communicating about the model and its training data with other team members or stakeholders. It provides a visual reference point for discussions and reviews.

Therefore, while the absence of a preview might not always indicate a critical error, it does remove a valuable feature for model validation and understanding. Resolving the missing preview issue ensures you have access to all the tools and information necessary for effective AI model management.

---

By systematically addressing the potential causes and following the recommended resolutions, you can effectively troubleshoot the issue of missing training document previews in AI Builder document processing models. Understanding the roles of permissions, ownership, and model import processes is key to maintaining and improving your AI-powered document automation solutions.

If you have encountered this issue, or have further questions or insights on troubleshooting missing training documents in AI Builder, please share your comments below! Your experiences can help other users in the community.