Document AI v1.4 foundation model no longer working

[GhassenB]

The pretrained foundation v1.4 model is returning empty extraction data without errors (not the case for previously finetuned extractors, or versions lower then 1.4).
Any explanation ? I can no longer fintune the v1.4 model nor use it in 0 shot 

[FoundationLab]

This started happening yesterday for me. I found the issue is with Foundation Models v1.4 and v1.5. If you roll back to v1.3 or you have a custom-trained model then things work again. This affects auto-labeling in the Google Cloud UI as well.

[GhassenB]

It's working again 🙂

[FoundationLab]

Thanks that's good to know. We already rolled our defaults back to 1.3 but I will try again now. I wonder if Google will ever acknowledge there was an issue, still nothing at https://status.cloud.google.com/products/GWuqLi6DKb1DkzyRtRuD/history.

[FoundationLab]

Final note: I see that v1.5 is "Google Stable" now (it was Release Candidate earlier this week). So I can only guess they hit a problem stabilizing v1.5 that somehow affected v1.4 too.