Hello,
We have been training an AI model with unstructured document types, and have run into some issues with model re-training as of a month ago. The following message has been coming up more and more frequently as we add additional training documents to the model, and is now failing more often than not.
"Training failed and the model cannot be used. Edit the model to retrain it. If the error persists, learn more about common issues​ and resolutions"
There doesn't appear to be much correlation between the successful and failed training phases from what we've seen so far. We've also confirmed that the documents are all under the maximum limits on this page (https://learn.microsoft.com/en-us/ai-builder/form-processing-model-requirements). There are 940 documents in this set (178 collections), and 5 fields are being extracted (2 of which are only present in a smaller percentage of the total dataset). All documents are readable PDFs (no handwriting).
Would anyone have any suggestions on what could be causing this issue? With the hefty dataset, I'm hoping we don't have to completely re-build the model from scratch, but would be grateful for any input on next steps here.
As discussed offline on aibhelp, some large files were causing trouble (many pages), so a workaround was to remove some files from the training set.
The number of training files for 'unstructured' shall remain small (few hundreds, not a thousand), and avoid when possible 'complex' documents with too many pages.
we're iterating on enhancing the reliability of this new 'unstructured' scenario.
Hello @wtfmt, When I'm reading this, I suggest you open a support ticket. You can do this by going to https://admin.powerplatform.microsoft.com
and click "Help + support"
_____________________________
Give a thumbs up if you like my answer.