Since roughly today, using the new model (non-legacy), we consistently get the following response for async vertex OCR tasks (DOCUMENT_TEXT_DETECTION) when processing PDFs with >5 pages:
{"error":{"code":3,"message":"Internal server error. Failed to process features."},"fullTextAnnotation":{},"context":{"uri":"scrubbed","pageNumber":6}}
We managed to get around this by switching to the old (legacy) model temporarily using the model: 'builtin/legacy' parameter, with the legacy model, everything works as expected.
We're using eu-vision.googleapis.com and making IBatchAnnotateFilesRequest using the JS SDK.
As there is no status for the Vision AI OCR service offering and the problem has been persisting for a couple of hours now, we are wondering whether this is just us or it is a known issue.
LinkedIn
Twitter
