While testing release (1.4.0) via Sbert I noticed that the Sentence Transformer was not processing Text if the OCR was being at the same time (so a new ingest) but was properly working when the OCR was already there and run unforced for the first time via Sbert.
The reason is the
$io->input->{$input_property} is an Object on the later but an array on the former. This is because of how serialization happens when already stored inside a keynote provider v/s when injected while-being-created from one Processor to another into a Queue worker.
Even if desired idea would be to normalize it might be actually not an easy task, so better to deal with that difference not at storage/retrieval time but on "connecting processors" time.
DiegoPino
commented
1 month ago
What?
While testing release (1.4.0) via Sbert I noticed that the Sentence Transformer was not processing Text if the OCR was being at the same time (so a new ingest) but was properly working when the OCR was already there and run unforced for the first time via Sbert.
The reason is the
$io->input->{$input_property}is an Object on the later but an array on the former. This is because of how serialization happens when already stored inside a keynote provider v/s when injected while-being-created from one Processor to another into a Queue worker.Even if desired idea would be to normalize it might be actually not an easy task, so better to deal with that difference not at storage/retrieval time but on "connecting processors" time.