unilm
unilm copied to clipboard
[TrOCR] How to train TRocr on custom dataset of different language
Hi, I want to change the Decoder part of TRocr to train and infer on different vocabulary [i.e different language]. I was following sample implementation from here but I was not able to change the vocabulary but just the size of vocabulary. It would be really helpful to know how can I change vocabulary for the TRocr decoder part.
Thanks
+1
hope to get some feedback.