CER exploding
#4
by davaa33 - opened
hello Im trying to finetune TrOCR for Mongolian Traditional Script and updated its tokenizer with SentencePiece tokenizer and started finetuning and after 1-4 Epochs depending on learning it starts to produce repeat tokens and CER just Explodes am I doing something wrong or is it just nature of a model ?
davaa33 changed discussion status to closed
if this is happening to you , lower your Transformer's version